2 * NET3 Protocol independent device support routines.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public License
6 * as published by the Free Software Foundation; either version
7 * 2 of the License, or (at your option) any later version.
9 * Derived from the non IP parts of dev.c 1.0.19
11 * Fred N. van Kempen, <waltje@uWalt.NL.Mugnet.ORG>
12 * Mark Evans, <evansmp@uhura.aston.ac.uk>
15 * Florian la Roche <rzsfl@rz.uni-sb.de>
16 * Alan Cox <gw4pts@gw4pts.ampr.org>
17 * David Hinds <dahinds@users.sourceforge.net>
18 * Alexey Kuznetsov <kuznet@ms2.inr.ac.ru>
19 * Adam Sulmicki <adam@cfar.umd.edu>
20 * Pekka Riikonen <priikone@poesidon.pspt.fi>
23 * D.J. Barrow : Fixed bug where dev->refcnt gets set
24 * to 2 if register_netdev gets called
25 * before net_dev_init & also removed a
26 * few lines of code in the process.
27 * Alan Cox : device private ioctl copies fields back.
28 * Alan Cox : Transmit queue code does relevant
29 * stunts to keep the queue safe.
30 * Alan Cox : Fixed double lock.
31 * Alan Cox : Fixed promisc NULL pointer trap
32 * ???????? : Support the full private ioctl range
33 * Alan Cox : Moved ioctl permission check into
35 * Tim Kordas : SIOCADDMULTI/SIOCDELMULTI
36 * Alan Cox : 100 backlog just doesn't cut it when
37 * you start doing multicast video 8)
38 * Alan Cox : Rewrote net_bh and list manager.
39 * Alan Cox : Fix ETH_P_ALL echoback lengths.
40 * Alan Cox : Took out transmit every packet pass
41 * Saved a few bytes in the ioctl handler
42 * Alan Cox : Network driver sets packet type before
43 * calling netif_rx. Saves a function
45 * Alan Cox : Hashed net_bh()
46 * Richard Kooijman: Timestamp fixes.
47 * Alan Cox : Wrong field in SIOCGIFDSTADDR
48 * Alan Cox : Device lock protection.
49 * Alan Cox : Fixed nasty side effect of device close
51 * Rudi Cilibrasi : Pass the right thing to
53 * Dave Miller : 32bit quantity for the device lock to
54 * make it work out on a Sparc.
55 * Bjorn Ekwall : Added KERNELD hack.
56 * Alan Cox : Cleaned up the backlog initialise.
57 * Craig Metz : SIOCGIFCONF fix if space for under
59 * Thomas Bogendoerfer : Return ENODEV for dev_open, if there
60 * is no device open function.
61 * Andi Kleen : Fix error reporting for SIOCGIFCONF
62 * Michael Chastain : Fix signed/unsigned for SIOCGIFCONF
63 * Cyrus Durgin : Cleaned for KMOD
64 * Adam Sulmicki : Bug Fix : Network Device Unload
65 * A network device unload needs to purge
67 * Paul Rusty Russell : SIOCSIFNAME
68 * Pekka Riikonen : Netdev boot-time settings code
69 * Andrew Morton : Make unregister_netdevice wait
70 * indefinitely on dev->refcnt
71 * J Hadi Salim : - Backlog queue sampling
72 * - netif_rx() feedback
75 #include <asm/uaccess.h>
76 #include <linux/bitops.h>
77 #include <linux/capability.h>
78 #include <linux/cpu.h>
79 #include <linux/types.h>
80 #include <linux/kernel.h>
81 #include <linux/hash.h>
82 #include <linux/slab.h>
83 #include <linux/sched.h>
84 #include <linux/mutex.h>
85 #include <linux/string.h>
87 #include <linux/socket.h>
88 #include <linux/sockios.h>
89 #include <linux/errno.h>
90 #include <linux/interrupt.h>
91 #include <linux/if_ether.h>
92 #include <linux/netdevice.h>
93 #include <linux/etherdevice.h>
94 #include <linux/ethtool.h>
95 #include <linux/notifier.h>
96 #include <linux/skbuff.h>
97 #include <net/net_namespace.h>
99 #include <linux/rtnetlink.h>
100 #include <linux/proc_fs.h>
101 #include <linux/seq_file.h>
102 #include <linux/stat.h>
104 #include <net/pkt_sched.h>
105 #include <net/checksum.h>
106 #include <net/xfrm.h>
107 #include <linux/highmem.h>
108 #include <linux/init.h>
109 #include <linux/kmod.h>
110 #include <linux/module.h>
111 #include <linux/netpoll.h>
112 #include <linux/rcupdate.h>
113 #include <linux/delay.h>
114 #include <net/wext.h>
115 #include <net/iw_handler.h>
116 #include <asm/current.h>
117 #include <linux/audit.h>
118 #include <linux/dmaengine.h>
119 #include <linux/err.h>
120 #include <linux/ctype.h>
121 #include <linux/if_arp.h>
122 #include <linux/if_vlan.h>
123 #include <linux/ip.h>
125 #include <linux/ipv6.h>
126 #include <linux/in.h>
127 #include <linux/jhash.h>
128 #include <linux/random.h>
129 #include <trace/events/napi.h>
130 #include <trace/events/net.h>
131 #include <trace/events/skb.h>
132 #include <linux/pci.h>
133 #include <linux/inetdevice.h>
134 #include <linux/cpu_rmap.h>
135 #include <linux/net_tstamp.h>
136 #include <linux/static_key.h>
137 #include <net/flow_keys.h>
139 #include "net-sysfs.h"
141 /* Instead of increasing this, you should create a hash table. */
142 #define MAX_GRO_SKBS 8
144 /* This should be increased if a protocol with a bigger head is added. */
145 #define GRO_MAX_HEAD (MAX_HEADER + 128)
148 * The list of packet types we will receive (as opposed to discard)
149 * and the routines to invoke.
151 * Why 16. Because with 16 the only overlap we get on a hash of the
152 * low nibble of the protocol value is RARP/SNAP/X.25.
154 * NOTE: That is no longer true with the addition of VLAN tags. Not
155 * sure which should go first, but I bet it won't make much
156 * difference if we are running VLANs. The good news is that
157 * this protocol won't be in the list unless compiled in, so
158 * the average user (w/out VLANs) will not be adversely affected.
175 #define PTYPE_HASH_SIZE (16)
176 #define PTYPE_HASH_MASK (PTYPE_HASH_SIZE - 1)
178 static DEFINE_SPINLOCK(ptype_lock
);
179 static struct list_head ptype_base
[PTYPE_HASH_SIZE
] __read_mostly
;
180 static struct list_head ptype_all __read_mostly
; /* Taps */
183 * The @dev_base_head list is protected by @dev_base_lock and the rtnl
186 * Pure readers hold dev_base_lock for reading, or rcu_read_lock()
188 * Writers must hold the rtnl semaphore while they loop through the
189 * dev_base_head list, and hold dev_base_lock for writing when they do the
190 * actual updates. This allows pure readers to access the list even
191 * while a writer is preparing to update it.
193 * To put it another way, dev_base_lock is held for writing only to
194 * protect against pure readers; the rtnl semaphore provides the
195 * protection against other writers.
197 * See, for example usages, register_netdevice() and
198 * unregister_netdevice(), which must be called with the rtnl
201 DEFINE_RWLOCK(dev_base_lock
);
202 EXPORT_SYMBOL(dev_base_lock
);
204 static inline void dev_base_seq_inc(struct net
*net
)
206 while (++net
->dev_base_seq
== 0);
209 static inline struct hlist_head
*dev_name_hash(struct net
*net
, const char *name
)
211 unsigned int hash
= full_name_hash(name
, strnlen(name
, IFNAMSIZ
));
213 return &net
->dev_name_head
[hash_32(hash
, NETDEV_HASHBITS
)];
216 static inline struct hlist_head
*dev_index_hash(struct net
*net
, int ifindex
)
218 return &net
->dev_index_head
[ifindex
& (NETDEV_HASHENTRIES
- 1)];
221 static inline void rps_lock(struct softnet_data
*sd
)
224 spin_lock(&sd
->input_pkt_queue
.lock
);
228 static inline void rps_unlock(struct softnet_data
*sd
)
231 spin_unlock(&sd
->input_pkt_queue
.lock
);
235 /* Device list insertion */
236 static int list_netdevice(struct net_device
*dev
)
238 struct net
*net
= dev_net(dev
);
242 write_lock_bh(&dev_base_lock
);
243 list_add_tail_rcu(&dev
->dev_list
, &net
->dev_base_head
);
244 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
245 hlist_add_head_rcu(&dev
->index_hlist
,
246 dev_index_hash(net
, dev
->ifindex
));
247 write_unlock_bh(&dev_base_lock
);
249 dev_base_seq_inc(net
);
254 /* Device list removal
255 * caller must respect a RCU grace period before freeing/reusing dev
257 static void unlist_netdevice(struct net_device
*dev
)
261 /* Unlink dev from the device chain */
262 write_lock_bh(&dev_base_lock
);
263 list_del_rcu(&dev
->dev_list
);
264 hlist_del_rcu(&dev
->name_hlist
);
265 hlist_del_rcu(&dev
->index_hlist
);
266 write_unlock_bh(&dev_base_lock
);
268 dev_base_seq_inc(dev_net(dev
));
275 static RAW_NOTIFIER_HEAD(netdev_chain
);
278 * Device drivers call our routines to queue packets here. We empty the
279 * queue in the local softnet handler.
282 DEFINE_PER_CPU_ALIGNED(struct softnet_data
, softnet_data
);
283 EXPORT_PER_CPU_SYMBOL(softnet_data
);
285 #ifdef CONFIG_LOCKDEP
287 * register_netdevice() inits txq->_xmit_lock and sets lockdep class
288 * according to dev->type
290 static const unsigned short netdev_lock_type
[] =
291 {ARPHRD_NETROM
, ARPHRD_ETHER
, ARPHRD_EETHER
, ARPHRD_AX25
,
292 ARPHRD_PRONET
, ARPHRD_CHAOS
, ARPHRD_IEEE802
, ARPHRD_ARCNET
,
293 ARPHRD_APPLETLK
, ARPHRD_DLCI
, ARPHRD_ATM
, ARPHRD_METRICOM
,
294 ARPHRD_IEEE1394
, ARPHRD_EUI64
, ARPHRD_INFINIBAND
, ARPHRD_SLIP
,
295 ARPHRD_CSLIP
, ARPHRD_SLIP6
, ARPHRD_CSLIP6
, ARPHRD_RSRVD
,
296 ARPHRD_ADAPT
, ARPHRD_ROSE
, ARPHRD_X25
, ARPHRD_HWX25
,
297 ARPHRD_PPP
, ARPHRD_CISCO
, ARPHRD_LAPB
, ARPHRD_DDCMP
,
298 ARPHRD_RAWHDLC
, ARPHRD_TUNNEL
, ARPHRD_TUNNEL6
, ARPHRD_FRAD
,
299 ARPHRD_SKIP
, ARPHRD_LOOPBACK
, ARPHRD_LOCALTLK
, ARPHRD_FDDI
,
300 ARPHRD_BIF
, ARPHRD_SIT
, ARPHRD_IPDDP
, ARPHRD_IPGRE
,
301 ARPHRD_PIMREG
, ARPHRD_HIPPI
, ARPHRD_ASH
, ARPHRD_ECONET
,
302 ARPHRD_IRDA
, ARPHRD_FCPP
, ARPHRD_FCAL
, ARPHRD_FCPL
,
303 ARPHRD_FCFABRIC
, ARPHRD_IEEE80211
, ARPHRD_IEEE80211_PRISM
,
304 ARPHRD_IEEE80211_RADIOTAP
, ARPHRD_PHONET
, ARPHRD_PHONET_PIPE
,
305 ARPHRD_IEEE802154
, ARPHRD_VOID
, ARPHRD_NONE
};
307 static const char *const netdev_lock_name
[] =
308 {"_xmit_NETROM", "_xmit_ETHER", "_xmit_EETHER", "_xmit_AX25",
309 "_xmit_PRONET", "_xmit_CHAOS", "_xmit_IEEE802", "_xmit_ARCNET",
310 "_xmit_APPLETLK", "_xmit_DLCI", "_xmit_ATM", "_xmit_METRICOM",
311 "_xmit_IEEE1394", "_xmit_EUI64", "_xmit_INFINIBAND", "_xmit_SLIP",
312 "_xmit_CSLIP", "_xmit_SLIP6", "_xmit_CSLIP6", "_xmit_RSRVD",
313 "_xmit_ADAPT", "_xmit_ROSE", "_xmit_X25", "_xmit_HWX25",
314 "_xmit_PPP", "_xmit_CISCO", "_xmit_LAPB", "_xmit_DDCMP",
315 "_xmit_RAWHDLC", "_xmit_TUNNEL", "_xmit_TUNNEL6", "_xmit_FRAD",
316 "_xmit_SKIP", "_xmit_LOOPBACK", "_xmit_LOCALTLK", "_xmit_FDDI",
317 "_xmit_BIF", "_xmit_SIT", "_xmit_IPDDP", "_xmit_IPGRE",
318 "_xmit_PIMREG", "_xmit_HIPPI", "_xmit_ASH", "_xmit_ECONET",
319 "_xmit_IRDA", "_xmit_FCPP", "_xmit_FCAL", "_xmit_FCPL",
320 "_xmit_FCFABRIC", "_xmit_IEEE80211", "_xmit_IEEE80211_PRISM",
321 "_xmit_IEEE80211_RADIOTAP", "_xmit_PHONET", "_xmit_PHONET_PIPE",
322 "_xmit_IEEE802154", "_xmit_VOID", "_xmit_NONE"};
324 static struct lock_class_key netdev_xmit_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
325 static struct lock_class_key netdev_addr_lock_key
[ARRAY_SIZE(netdev_lock_type
)];
327 static inline unsigned short netdev_lock_pos(unsigned short dev_type
)
331 for (i
= 0; i
< ARRAY_SIZE(netdev_lock_type
); i
++)
332 if (netdev_lock_type
[i
] == dev_type
)
334 /* the last key is used by default */
335 return ARRAY_SIZE(netdev_lock_type
) - 1;
338 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
339 unsigned short dev_type
)
343 i
= netdev_lock_pos(dev_type
);
344 lockdep_set_class_and_name(lock
, &netdev_xmit_lock_key
[i
],
345 netdev_lock_name
[i
]);
348 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
352 i
= netdev_lock_pos(dev
->type
);
353 lockdep_set_class_and_name(&dev
->addr_list_lock
,
354 &netdev_addr_lock_key
[i
],
355 netdev_lock_name
[i
]);
358 static inline void netdev_set_xmit_lockdep_class(spinlock_t
*lock
,
359 unsigned short dev_type
)
362 static inline void netdev_set_addr_lockdep_class(struct net_device
*dev
)
367 /*******************************************************************************
369 Protocol management and registration routines
371 *******************************************************************************/
374 * Add a protocol ID to the list. Now that the input handler is
375 * smarter we can dispense with all the messy stuff that used to be
378 * BEWARE!!! Protocol handlers, mangling input packets,
379 * MUST BE last in hash buckets and checking protocol handlers
380 * MUST start from promiscuous ptype_all chain in net_bh.
381 * It is true now, do not change it.
382 * Explanation follows: if protocol handler, mangling packet, will
383 * be the first on list, it is not able to sense, that packet
384 * is cloned and should be copied-on-write, so that it will
385 * change it and subsequent readers will get broken packet.
389 static inline struct list_head
*ptype_head(const struct packet_type
*pt
)
391 if (pt
->type
== htons(ETH_P_ALL
))
394 return &ptype_base
[ntohs(pt
->type
) & PTYPE_HASH_MASK
];
398 * dev_add_pack - add packet handler
399 * @pt: packet type declaration
401 * Add a protocol handler to the networking stack. The passed &packet_type
402 * is linked into kernel lists and may not be freed until it has been
403 * removed from the kernel lists.
405 * This call does not sleep therefore it can not
406 * guarantee all CPU's that are in middle of receiving packets
407 * will see the new packet type (until the next received packet).
410 void dev_add_pack(struct packet_type
*pt
)
412 struct list_head
*head
= ptype_head(pt
);
414 spin_lock(&ptype_lock
);
415 list_add_rcu(&pt
->list
, head
);
416 spin_unlock(&ptype_lock
);
418 EXPORT_SYMBOL(dev_add_pack
);
421 * __dev_remove_pack - remove packet handler
422 * @pt: packet type declaration
424 * Remove a protocol handler that was previously added to the kernel
425 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
426 * from the kernel lists and can be freed or reused once this function
429 * The packet type might still be in use by receivers
430 * and must not be freed until after all the CPU's have gone
431 * through a quiescent state.
433 void __dev_remove_pack(struct packet_type
*pt
)
435 struct list_head
*head
= ptype_head(pt
);
436 struct packet_type
*pt1
;
438 spin_lock(&ptype_lock
);
440 list_for_each_entry(pt1
, head
, list
) {
442 list_del_rcu(&pt
->list
);
447 pr_warn("dev_remove_pack: %p not found\n", pt
);
449 spin_unlock(&ptype_lock
);
451 EXPORT_SYMBOL(__dev_remove_pack
);
454 * dev_remove_pack - remove packet handler
455 * @pt: packet type declaration
457 * Remove a protocol handler that was previously added to the kernel
458 * protocol handlers by dev_add_pack(). The passed &packet_type is removed
459 * from the kernel lists and can be freed or reused once this function
462 * This call sleeps to guarantee that no CPU is looking at the packet
465 void dev_remove_pack(struct packet_type
*pt
)
467 __dev_remove_pack(pt
);
471 EXPORT_SYMBOL(dev_remove_pack
);
473 /******************************************************************************
475 Device Boot-time Settings Routines
477 *******************************************************************************/
479 /* Boot time configuration table */
480 static struct netdev_boot_setup dev_boot_setup
[NETDEV_BOOT_SETUP_MAX
];
483 * netdev_boot_setup_add - add new setup entry
484 * @name: name of the device
485 * @map: configured settings for the device
487 * Adds new setup entry to the dev_boot_setup list. The function
488 * returns 0 on error and 1 on success. This is a generic routine to
491 static int netdev_boot_setup_add(char *name
, struct ifmap
*map
)
493 struct netdev_boot_setup
*s
;
497 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
498 if (s
[i
].name
[0] == '\0' || s
[i
].name
[0] == ' ') {
499 memset(s
[i
].name
, 0, sizeof(s
[i
].name
));
500 strlcpy(s
[i
].name
, name
, IFNAMSIZ
);
501 memcpy(&s
[i
].map
, map
, sizeof(s
[i
].map
));
506 return i
>= NETDEV_BOOT_SETUP_MAX
? 0 : 1;
510 * netdev_boot_setup_check - check boot time settings
511 * @dev: the netdevice
513 * Check boot time settings for the device.
514 * The found settings are set for the device to be used
515 * later in the device probing.
516 * Returns 0 if no settings found, 1 if they are.
518 int netdev_boot_setup_check(struct net_device
*dev
)
520 struct netdev_boot_setup
*s
= dev_boot_setup
;
523 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++) {
524 if (s
[i
].name
[0] != '\0' && s
[i
].name
[0] != ' ' &&
525 !strcmp(dev
->name
, s
[i
].name
)) {
526 dev
->irq
= s
[i
].map
.irq
;
527 dev
->base_addr
= s
[i
].map
.base_addr
;
528 dev
->mem_start
= s
[i
].map
.mem_start
;
529 dev
->mem_end
= s
[i
].map
.mem_end
;
535 EXPORT_SYMBOL(netdev_boot_setup_check
);
539 * netdev_boot_base - get address from boot time settings
540 * @prefix: prefix for network device
541 * @unit: id for network device
543 * Check boot time settings for the base address of device.
544 * The found settings are set for the device to be used
545 * later in the device probing.
546 * Returns 0 if no settings found.
548 unsigned long netdev_boot_base(const char *prefix
, int unit
)
550 const struct netdev_boot_setup
*s
= dev_boot_setup
;
554 sprintf(name
, "%s%d", prefix
, unit
);
557 * If device already registered then return base of 1
558 * to indicate not to probe for this interface
560 if (__dev_get_by_name(&init_net
, name
))
563 for (i
= 0; i
< NETDEV_BOOT_SETUP_MAX
; i
++)
564 if (!strcmp(name
, s
[i
].name
))
565 return s
[i
].map
.base_addr
;
570 * Saves at boot time configured settings for any netdevice.
572 int __init
netdev_boot_setup(char *str
)
577 str
= get_options(str
, ARRAY_SIZE(ints
), ints
);
582 memset(&map
, 0, sizeof(map
));
586 map
.base_addr
= ints
[2];
588 map
.mem_start
= ints
[3];
590 map
.mem_end
= ints
[4];
592 /* Add new entry to the list */
593 return netdev_boot_setup_add(str
, &map
);
596 __setup("netdev=", netdev_boot_setup
);
598 /*******************************************************************************
600 Device Interface Subroutines
602 *******************************************************************************/
605 * __dev_get_by_name - find a device by its name
606 * @net: the applicable net namespace
607 * @name: name to find
609 * Find an interface by name. Must be called under RTNL semaphore
610 * or @dev_base_lock. If the name is found a pointer to the device
611 * is returned. If the name is not found then %NULL is returned. The
612 * reference counters are not incremented so the caller must be
613 * careful with locks.
616 struct net_device
*__dev_get_by_name(struct net
*net
, const char *name
)
618 struct hlist_node
*p
;
619 struct net_device
*dev
;
620 struct hlist_head
*head
= dev_name_hash(net
, name
);
622 hlist_for_each_entry(dev
, p
, head
, name_hlist
)
623 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
628 EXPORT_SYMBOL(__dev_get_by_name
);
631 * dev_get_by_name_rcu - find a device by its name
632 * @net: the applicable net namespace
633 * @name: name to find
635 * Find an interface by name.
636 * If the name is found a pointer to the device is returned.
637 * If the name is not found then %NULL is returned.
638 * The reference counters are not incremented so the caller must be
639 * careful with locks. The caller must hold RCU lock.
642 struct net_device
*dev_get_by_name_rcu(struct net
*net
, const char *name
)
644 struct hlist_node
*p
;
645 struct net_device
*dev
;
646 struct hlist_head
*head
= dev_name_hash(net
, name
);
648 hlist_for_each_entry_rcu(dev
, p
, head
, name_hlist
)
649 if (!strncmp(dev
->name
, name
, IFNAMSIZ
))
654 EXPORT_SYMBOL(dev_get_by_name_rcu
);
657 * dev_get_by_name - find a device by its name
658 * @net: the applicable net namespace
659 * @name: name to find
661 * Find an interface by name. This can be called from any
662 * context and does its own locking. The returned handle has
663 * the usage count incremented and the caller must use dev_put() to
664 * release it when it is no longer needed. %NULL is returned if no
665 * matching device is found.
668 struct net_device
*dev_get_by_name(struct net
*net
, const char *name
)
670 struct net_device
*dev
;
673 dev
= dev_get_by_name_rcu(net
, name
);
679 EXPORT_SYMBOL(dev_get_by_name
);
682 * __dev_get_by_index - find a device by its ifindex
683 * @net: the applicable net namespace
684 * @ifindex: index of device
686 * Search for an interface by index. Returns %NULL if the device
687 * is not found or a pointer to the device. The device has not
688 * had its reference counter increased so the caller must be careful
689 * about locking. The caller must hold either the RTNL semaphore
693 struct net_device
*__dev_get_by_index(struct net
*net
, int ifindex
)
695 struct hlist_node
*p
;
696 struct net_device
*dev
;
697 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
699 hlist_for_each_entry(dev
, p
, head
, index_hlist
)
700 if (dev
->ifindex
== ifindex
)
705 EXPORT_SYMBOL(__dev_get_by_index
);
708 * dev_get_by_index_rcu - find a device by its ifindex
709 * @net: the applicable net namespace
710 * @ifindex: index of device
712 * Search for an interface by index. Returns %NULL if the device
713 * is not found or a pointer to the device. The device has not
714 * had its reference counter increased so the caller must be careful
715 * about locking. The caller must hold RCU lock.
718 struct net_device
*dev_get_by_index_rcu(struct net
*net
, int ifindex
)
720 struct hlist_node
*p
;
721 struct net_device
*dev
;
722 struct hlist_head
*head
= dev_index_hash(net
, ifindex
);
724 hlist_for_each_entry_rcu(dev
, p
, head
, index_hlist
)
725 if (dev
->ifindex
== ifindex
)
730 EXPORT_SYMBOL(dev_get_by_index_rcu
);
734 * dev_get_by_index - find a device by its ifindex
735 * @net: the applicable net namespace
736 * @ifindex: index of device
738 * Search for an interface by index. Returns NULL if the device
739 * is not found or a pointer to the device. The device returned has
740 * had a reference added and the pointer is safe until the user calls
741 * dev_put to indicate they have finished with it.
744 struct net_device
*dev_get_by_index(struct net
*net
, int ifindex
)
746 struct net_device
*dev
;
749 dev
= dev_get_by_index_rcu(net
, ifindex
);
755 EXPORT_SYMBOL(dev_get_by_index
);
758 * dev_getbyhwaddr_rcu - find a device by its hardware address
759 * @net: the applicable net namespace
760 * @type: media type of device
761 * @ha: hardware address
763 * Search for an interface by MAC address. Returns NULL if the device
764 * is not found or a pointer to the device.
765 * The caller must hold RCU or RTNL.
766 * The returned device has not had its ref count increased
767 * and the caller must therefore be careful about locking
771 struct net_device
*dev_getbyhwaddr_rcu(struct net
*net
, unsigned short type
,
774 struct net_device
*dev
;
776 for_each_netdev_rcu(net
, dev
)
777 if (dev
->type
== type
&&
778 !memcmp(dev
->dev_addr
, ha
, dev
->addr_len
))
783 EXPORT_SYMBOL(dev_getbyhwaddr_rcu
);
785 struct net_device
*__dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
787 struct net_device
*dev
;
790 for_each_netdev(net
, dev
)
791 if (dev
->type
== type
)
796 EXPORT_SYMBOL(__dev_getfirstbyhwtype
);
798 struct net_device
*dev_getfirstbyhwtype(struct net
*net
, unsigned short type
)
800 struct net_device
*dev
, *ret
= NULL
;
803 for_each_netdev_rcu(net
, dev
)
804 if (dev
->type
== type
) {
812 EXPORT_SYMBOL(dev_getfirstbyhwtype
);
815 * dev_get_by_flags_rcu - find any device with given flags
816 * @net: the applicable net namespace
817 * @if_flags: IFF_* values
818 * @mask: bitmask of bits in if_flags to check
820 * Search for any interface with the given flags. Returns NULL if a device
821 * is not found or a pointer to the device. Must be called inside
822 * rcu_read_lock(), and result refcount is unchanged.
825 struct net_device
*dev_get_by_flags_rcu(struct net
*net
, unsigned short if_flags
,
828 struct net_device
*dev
, *ret
;
831 for_each_netdev_rcu(net
, dev
) {
832 if (((dev
->flags
^ if_flags
) & mask
) == 0) {
839 EXPORT_SYMBOL(dev_get_by_flags_rcu
);
842 * dev_valid_name - check if name is okay for network device
845 * Network device names need to be valid file names to
846 * to allow sysfs to work. We also disallow any kind of
849 bool dev_valid_name(const char *name
)
853 if (strlen(name
) >= IFNAMSIZ
)
855 if (!strcmp(name
, ".") || !strcmp(name
, ".."))
859 if (*name
== '/' || isspace(*name
))
865 EXPORT_SYMBOL(dev_valid_name
);
868 * __dev_alloc_name - allocate a name for a device
869 * @net: network namespace to allocate the device name in
870 * @name: name format string
871 * @buf: scratch buffer and result name string
873 * Passed a format string - eg "lt%d" it will try and find a suitable
874 * id. It scans list of devices to build up a free map, then chooses
875 * the first empty slot. The caller must hold the dev_base or rtnl lock
876 * while allocating the name and adding the device in order to avoid
878 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
879 * Returns the number of the unit assigned or a negative errno code.
882 static int __dev_alloc_name(struct net
*net
, const char *name
, char *buf
)
886 const int max_netdevices
= 8*PAGE_SIZE
;
887 unsigned long *inuse
;
888 struct net_device
*d
;
890 p
= strnchr(name
, IFNAMSIZ
-1, '%');
893 * Verify the string as this thing may have come from
894 * the user. There must be either one "%d" and no other "%"
897 if (p
[1] != 'd' || strchr(p
+ 2, '%'))
900 /* Use one page as a bit array of possible slots */
901 inuse
= (unsigned long *) get_zeroed_page(GFP_ATOMIC
);
905 for_each_netdev(net
, d
) {
906 if (!sscanf(d
->name
, name
, &i
))
908 if (i
< 0 || i
>= max_netdevices
)
911 /* avoid cases where sscanf is not exact inverse of printf */
912 snprintf(buf
, IFNAMSIZ
, name
, i
);
913 if (!strncmp(buf
, d
->name
, IFNAMSIZ
))
917 i
= find_first_zero_bit(inuse
, max_netdevices
);
918 free_page((unsigned long) inuse
);
922 snprintf(buf
, IFNAMSIZ
, name
, i
);
923 if (!__dev_get_by_name(net
, buf
))
926 /* It is possible to run out of possible slots
927 * when the name is long and there isn't enough space left
928 * for the digits, or if all bits are used.
934 * dev_alloc_name - allocate a name for a device
936 * @name: name format string
938 * Passed a format string - eg "lt%d" it will try and find a suitable
939 * id. It scans list of devices to build up a free map, then chooses
940 * the first empty slot. The caller must hold the dev_base or rtnl lock
941 * while allocating the name and adding the device in order to avoid
943 * Limited to bits_per_byte * page size devices (ie 32K on most platforms).
944 * Returns the number of the unit assigned or a negative errno code.
947 int dev_alloc_name(struct net_device
*dev
, const char *name
)
953 BUG_ON(!dev_net(dev
));
955 ret
= __dev_alloc_name(net
, name
, buf
);
957 strlcpy(dev
->name
, buf
, IFNAMSIZ
);
960 EXPORT_SYMBOL(dev_alloc_name
);
962 static int dev_get_valid_name(struct net_device
*dev
, const char *name
)
966 BUG_ON(!dev_net(dev
));
969 if (!dev_valid_name(name
))
972 if (strchr(name
, '%'))
973 return dev_alloc_name(dev
, name
);
974 else if (__dev_get_by_name(net
, name
))
976 else if (dev
->name
!= name
)
977 strlcpy(dev
->name
, name
, IFNAMSIZ
);
983 * dev_change_name - change name of a device
985 * @newname: name (or format string) must be at least IFNAMSIZ
987 * Change name of a device, can pass format strings "eth%d".
990 int dev_change_name(struct net_device
*dev
, const char *newname
)
992 char oldname
[IFNAMSIZ
];
998 BUG_ON(!dev_net(dev
));
1001 if (dev
->flags
& IFF_UP
)
1004 if (strncmp(newname
, dev
->name
, IFNAMSIZ
) == 0)
1007 memcpy(oldname
, dev
->name
, IFNAMSIZ
);
1009 err
= dev_get_valid_name(dev
, newname
);
1014 ret
= device_rename(&dev
->dev
, dev
->name
);
1016 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1020 write_lock_bh(&dev_base_lock
);
1021 hlist_del_rcu(&dev
->name_hlist
);
1022 write_unlock_bh(&dev_base_lock
);
1026 write_lock_bh(&dev_base_lock
);
1027 hlist_add_head_rcu(&dev
->name_hlist
, dev_name_hash(net
, dev
->name
));
1028 write_unlock_bh(&dev_base_lock
);
1030 ret
= call_netdevice_notifiers(NETDEV_CHANGENAME
, dev
);
1031 ret
= notifier_to_errno(ret
);
1034 /* err >= 0 after dev_alloc_name() or stores the first errno */
1037 memcpy(dev
->name
, oldname
, IFNAMSIZ
);
1040 pr_err("%s: name change rollback failed: %d\n",
1049 * dev_set_alias - change ifalias of a device
1051 * @alias: name up to IFALIASZ
1052 * @len: limit of bytes to copy from info
1054 * Set ifalias for a device,
1056 int dev_set_alias(struct net_device
*dev
, const char *alias
, size_t len
)
1060 if (len
>= IFALIASZ
)
1065 kfree(dev
->ifalias
);
1066 dev
->ifalias
= NULL
;
1071 dev
->ifalias
= krealloc(dev
->ifalias
, len
+ 1, GFP_KERNEL
);
1075 strlcpy(dev
->ifalias
, alias
, len
+1);
1081 * netdev_features_change - device changes features
1082 * @dev: device to cause notification
1084 * Called to indicate a device has changed features.
1086 void netdev_features_change(struct net_device
*dev
)
1088 call_netdevice_notifiers(NETDEV_FEAT_CHANGE
, dev
);
1090 EXPORT_SYMBOL(netdev_features_change
);
1093 * netdev_state_change - device changes state
1094 * @dev: device to cause notification
1096 * Called to indicate a device has changed state. This function calls
1097 * the notifier chains for netdev_chain and sends a NEWLINK message
1098 * to the routing socket.
1100 void netdev_state_change(struct net_device
*dev
)
1102 if (dev
->flags
& IFF_UP
) {
1103 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
1104 rtmsg_ifinfo(RTM_NEWLINK
, dev
, 0);
1107 EXPORT_SYMBOL(netdev_state_change
);
1109 int netdev_bonding_change(struct net_device
*dev
, unsigned long event
)
1111 return call_netdevice_notifiers(event
, dev
);
1113 EXPORT_SYMBOL(netdev_bonding_change
);
1116 * dev_load - load a network module
1117 * @net: the applicable net namespace
1118 * @name: name of interface
1120 * If a network interface is not present and the process has suitable
1121 * privileges this function loads the module. If module loading is not
1122 * available in this kernel then it becomes a nop.
1125 void dev_load(struct net
*net
, const char *name
)
1127 struct net_device
*dev
;
1131 dev
= dev_get_by_name_rcu(net
, name
);
1135 if (no_module
&& capable(CAP_NET_ADMIN
))
1136 no_module
= request_module("netdev-%s", name
);
1137 if (no_module
&& capable(CAP_SYS_MODULE
)) {
1138 if (!request_module("%s", name
))
1139 pr_warn("Loading kernel module for a network device with CAP_SYS_MODULE (deprecated). Use CAP_NET_ADMIN and alias netdev-%s instead.\n",
1143 EXPORT_SYMBOL(dev_load
);
1145 static int __dev_open(struct net_device
*dev
)
1147 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1152 if (!netif_device_present(dev
))
1155 ret
= call_netdevice_notifiers(NETDEV_PRE_UP
, dev
);
1156 ret
= notifier_to_errno(ret
);
1160 set_bit(__LINK_STATE_START
, &dev
->state
);
1162 if (ops
->ndo_validate_addr
)
1163 ret
= ops
->ndo_validate_addr(dev
);
1165 if (!ret
&& ops
->ndo_open
)
1166 ret
= ops
->ndo_open(dev
);
1169 clear_bit(__LINK_STATE_START
, &dev
->state
);
1171 dev
->flags
|= IFF_UP
;
1172 net_dmaengine_get();
1173 dev_set_rx_mode(dev
);
1181 * dev_open - prepare an interface for use.
1182 * @dev: device to open
1184 * Takes a device from down to up state. The device's private open
1185 * function is invoked and then the multicast lists are loaded. Finally
1186 * the device is moved into the up state and a %NETDEV_UP message is
1187 * sent to the netdev notifier chain.
1189 * Calling this function on an active interface is a nop. On a failure
1190 * a negative errno code is returned.
1192 int dev_open(struct net_device
*dev
)
1196 if (dev
->flags
& IFF_UP
)
1199 ret
= __dev_open(dev
);
1203 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1204 call_netdevice_notifiers(NETDEV_UP
, dev
);
1208 EXPORT_SYMBOL(dev_open
);
1210 static int __dev_close_many(struct list_head
*head
)
1212 struct net_device
*dev
;
1217 list_for_each_entry(dev
, head
, unreg_list
) {
1218 call_netdevice_notifiers(NETDEV_GOING_DOWN
, dev
);
1220 clear_bit(__LINK_STATE_START
, &dev
->state
);
1222 /* Synchronize to scheduled poll. We cannot touch poll list, it
1223 * can be even on different cpu. So just clear netif_running().
1225 * dev->stop() will invoke napi_disable() on all of it's
1226 * napi_struct instances on this device.
1228 smp_mb__after_clear_bit(); /* Commit netif_running(). */
1231 dev_deactivate_many(head
);
1233 list_for_each_entry(dev
, head
, unreg_list
) {
1234 const struct net_device_ops
*ops
= dev
->netdev_ops
;
1237 * Call the device specific close. This cannot fail.
1238 * Only if device is UP
1240 * We allow it to be called even after a DETACH hot-plug
1246 dev
->flags
&= ~IFF_UP
;
1247 net_dmaengine_put();
1253 static int __dev_close(struct net_device
*dev
)
1258 list_add(&dev
->unreg_list
, &single
);
1259 retval
= __dev_close_many(&single
);
1264 static int dev_close_many(struct list_head
*head
)
1266 struct net_device
*dev
, *tmp
;
1267 LIST_HEAD(tmp_list
);
1269 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
)
1270 if (!(dev
->flags
& IFF_UP
))
1271 list_move(&dev
->unreg_list
, &tmp_list
);
1273 __dev_close_many(head
);
1275 list_for_each_entry(dev
, head
, unreg_list
) {
1276 rtmsg_ifinfo(RTM_NEWLINK
, dev
, IFF_UP
|IFF_RUNNING
);
1277 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
1280 /* rollback_registered_many needs the complete original list */
1281 list_splice(&tmp_list
, head
);
1286 * dev_close - shutdown an interface.
1287 * @dev: device to shutdown
1289 * This function moves an active device into down state. A
1290 * %NETDEV_GOING_DOWN is sent to the netdev notifier chain. The device
1291 * is then deactivated and finally a %NETDEV_DOWN is sent to the notifier
1294 int dev_close(struct net_device
*dev
)
1296 if (dev
->flags
& IFF_UP
) {
1299 list_add(&dev
->unreg_list
, &single
);
1300 dev_close_many(&single
);
1305 EXPORT_SYMBOL(dev_close
);
1309 * dev_disable_lro - disable Large Receive Offload on a device
1312 * Disable Large Receive Offload (LRO) on a net device. Must be
1313 * called under RTNL. This is needed if received packets may be
1314 * forwarded to another interface.
1316 void dev_disable_lro(struct net_device
*dev
)
1319 * If we're trying to disable lro on a vlan device
1320 * use the underlying physical device instead
1322 if (is_vlan_dev(dev
))
1323 dev
= vlan_dev_real_dev(dev
);
1325 dev
->wanted_features
&= ~NETIF_F_LRO
;
1326 netdev_update_features(dev
);
1328 if (unlikely(dev
->features
& NETIF_F_LRO
))
1329 netdev_WARN(dev
, "failed to disable LRO!\n");
1331 EXPORT_SYMBOL(dev_disable_lro
);
1334 static int dev_boot_phase
= 1;
1337 * register_netdevice_notifier - register a network notifier block
1340 * Register a notifier to be called when network device events occur.
1341 * The notifier passed is linked into the kernel structures and must
1342 * not be reused until it has been unregistered. A negative errno code
1343 * is returned on a failure.
1345 * When registered all registration and up events are replayed
1346 * to the new notifier to allow device to have a race free
1347 * view of the network device list.
1350 int register_netdevice_notifier(struct notifier_block
*nb
)
1352 struct net_device
*dev
;
1353 struct net_device
*last
;
1358 err
= raw_notifier_chain_register(&netdev_chain
, nb
);
1364 for_each_netdev(net
, dev
) {
1365 err
= nb
->notifier_call(nb
, NETDEV_REGISTER
, dev
);
1366 err
= notifier_to_errno(err
);
1370 if (!(dev
->flags
& IFF_UP
))
1373 nb
->notifier_call(nb
, NETDEV_UP
, dev
);
1384 for_each_netdev(net
, dev
) {
1388 if (dev
->flags
& IFF_UP
) {
1389 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1390 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1392 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1393 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1398 raw_notifier_chain_unregister(&netdev_chain
, nb
);
1401 EXPORT_SYMBOL(register_netdevice_notifier
);
1404 * unregister_netdevice_notifier - unregister a network notifier block
1407 * Unregister a notifier previously registered by
1408 * register_netdevice_notifier(). The notifier is unlinked into the
1409 * kernel structures and may then be reused. A negative errno code
1410 * is returned on a failure.
1412 * After unregistering unregister and down device events are synthesized
1413 * for all devices on the device list to the removed notifier to remove
1414 * the need for special case cleanup code.
1417 int unregister_netdevice_notifier(struct notifier_block
*nb
)
1419 struct net_device
*dev
;
1424 err
= raw_notifier_chain_unregister(&netdev_chain
, nb
);
1429 for_each_netdev(net
, dev
) {
1430 if (dev
->flags
& IFF_UP
) {
1431 nb
->notifier_call(nb
, NETDEV_GOING_DOWN
, dev
);
1432 nb
->notifier_call(nb
, NETDEV_DOWN
, dev
);
1434 nb
->notifier_call(nb
, NETDEV_UNREGISTER
, dev
);
1435 nb
->notifier_call(nb
, NETDEV_UNREGISTER_BATCH
, dev
);
1442 EXPORT_SYMBOL(unregister_netdevice_notifier
);
1445 * call_netdevice_notifiers - call all network notifier blocks
1446 * @val: value passed unmodified to notifier function
1447 * @dev: net_device pointer passed unmodified to notifier function
1449 * Call all network notifier blocks. Parameters and return value
1450 * are as for raw_notifier_call_chain().
1453 int call_netdevice_notifiers(unsigned long val
, struct net_device
*dev
)
1456 return raw_notifier_call_chain(&netdev_chain
, val
, dev
);
1458 EXPORT_SYMBOL(call_netdevice_notifiers
);
1460 static struct static_key netstamp_needed __read_mostly
;
1461 #ifdef HAVE_JUMP_LABEL
1462 /* We are not allowed to call static_key_slow_dec() from irq context
1463 * If net_disable_timestamp() is called from irq context, defer the
1464 * static_key_slow_dec() calls.
1466 static atomic_t netstamp_needed_deferred
;
1469 void net_enable_timestamp(void)
1471 #ifdef HAVE_JUMP_LABEL
1472 int deferred
= atomic_xchg(&netstamp_needed_deferred
, 0);
1476 static_key_slow_dec(&netstamp_needed
);
1480 WARN_ON(in_interrupt());
1481 static_key_slow_inc(&netstamp_needed
);
1483 EXPORT_SYMBOL(net_enable_timestamp
);
1485 void net_disable_timestamp(void)
1487 #ifdef HAVE_JUMP_LABEL
1488 if (in_interrupt()) {
1489 atomic_inc(&netstamp_needed_deferred
);
1493 static_key_slow_dec(&netstamp_needed
);
1495 EXPORT_SYMBOL(net_disable_timestamp
);
1497 static inline void net_timestamp_set(struct sk_buff
*skb
)
1499 skb
->tstamp
.tv64
= 0;
1500 if (static_key_false(&netstamp_needed
))
1501 __net_timestamp(skb
);
1504 #define net_timestamp_check(COND, SKB) \
1505 if (static_key_false(&netstamp_needed)) { \
1506 if ((COND) && !(SKB)->tstamp.tv64) \
1507 __net_timestamp(SKB); \
1510 static int net_hwtstamp_validate(struct ifreq *ifr)
1512 struct hwtstamp_config cfg
;
1513 enum hwtstamp_tx_types tx_type
;
1514 enum hwtstamp_rx_filters rx_filter
;
1515 int tx_type_valid
= 0;
1516 int rx_filter_valid
= 0;
1518 if (copy_from_user(&cfg
, ifr
->ifr_data
, sizeof(cfg
)))
1521 if (cfg
.flags
) /* reserved for future extensions */
1524 tx_type
= cfg
.tx_type
;
1525 rx_filter
= cfg
.rx_filter
;
1528 case HWTSTAMP_TX_OFF
:
1529 case HWTSTAMP_TX_ON
:
1530 case HWTSTAMP_TX_ONESTEP_SYNC
:
1535 switch (rx_filter
) {
1536 case HWTSTAMP_FILTER_NONE
:
1537 case HWTSTAMP_FILTER_ALL
:
1538 case HWTSTAMP_FILTER_SOME
:
1539 case HWTSTAMP_FILTER_PTP_V1_L4_EVENT
:
1540 case HWTSTAMP_FILTER_PTP_V1_L4_SYNC
:
1541 case HWTSTAMP_FILTER_PTP_V1_L4_DELAY_REQ
:
1542 case HWTSTAMP_FILTER_PTP_V2_L4_EVENT
:
1543 case HWTSTAMP_FILTER_PTP_V2_L4_SYNC
:
1544 case HWTSTAMP_FILTER_PTP_V2_L4_DELAY_REQ
:
1545 case HWTSTAMP_FILTER_PTP_V2_L2_EVENT
:
1546 case HWTSTAMP_FILTER_PTP_V2_L2_SYNC
:
1547 case HWTSTAMP_FILTER_PTP_V2_L2_DELAY_REQ
:
1548 case HWTSTAMP_FILTER_PTP_V2_EVENT
:
1549 case HWTSTAMP_FILTER_PTP_V2_SYNC
:
1550 case HWTSTAMP_FILTER_PTP_V2_DELAY_REQ
:
1551 rx_filter_valid
= 1;
1555 if (!tx_type_valid
|| !rx_filter_valid
)
1561 static inline bool is_skb_forwardable(struct net_device
*dev
,
1562 struct sk_buff
*skb
)
1566 if (!(dev
->flags
& IFF_UP
))
1569 len
= dev
->mtu
+ dev
->hard_header_len
+ VLAN_HLEN
;
1570 if (skb
->len
<= len
)
1573 /* if TSO is enabled, we don't care about the length as the packet
1574 * could be forwarded without being segmented before
1576 if (skb_is_gso(skb
))
1583 * dev_forward_skb - loopback an skb to another netif
1585 * @dev: destination network device
1586 * @skb: buffer to forward
1589 * NET_RX_SUCCESS (no congestion)
1590 * NET_RX_DROP (packet was dropped, but freed)
1592 * dev_forward_skb can be used for injecting an skb from the
1593 * start_xmit function of one device into the receive queue
1594 * of another device.
1596 * The receiving device may be in another namespace, so
1597 * we have to clear all information in the skb that could
1598 * impact namespace isolation.
1600 int dev_forward_skb(struct net_device
*dev
, struct sk_buff
*skb
)
1602 if (skb_shinfo(skb
)->tx_flags
& SKBTX_DEV_ZEROCOPY
) {
1603 if (skb_copy_ubufs(skb
, GFP_ATOMIC
)) {
1604 atomic_long_inc(&dev
->rx_dropped
);
1613 if (unlikely(!is_skb_forwardable(dev
, skb
))) {
1614 atomic_long_inc(&dev
->rx_dropped
);
1621 skb
->tstamp
.tv64
= 0;
1622 skb
->pkt_type
= PACKET_HOST
;
1623 skb
->protocol
= eth_type_trans(skb
, dev
);
1627 return netif_rx(skb
);
1629 EXPORT_SYMBOL_GPL(dev_forward_skb
);
1631 static inline int deliver_skb(struct sk_buff
*skb
,
1632 struct packet_type
*pt_prev
,
1633 struct net_device
*orig_dev
)
1635 atomic_inc(&skb
->users
);
1636 return pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
1640 * Support routine. Sends outgoing frames to any network
1641 * taps currently in use.
1644 static void dev_queue_xmit_nit(struct sk_buff
*skb
, struct net_device
*dev
)
1646 struct packet_type
*ptype
;
1647 struct sk_buff
*skb2
= NULL
;
1648 struct packet_type
*pt_prev
= NULL
;
1651 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
1652 /* Never send packets back to the socket
1653 * they originated from - MvS (miquels@drinkel.ow.org)
1655 if ((ptype
->dev
== dev
|| !ptype
->dev
) &&
1656 (ptype
->af_packet_priv
== NULL
||
1657 (struct sock
*)ptype
->af_packet_priv
!= skb
->sk
)) {
1659 deliver_skb(skb2
, pt_prev
, skb
->dev
);
1664 skb2
= skb_clone(skb
, GFP_ATOMIC
);
1668 net_timestamp_set(skb2
);
1670 /* skb->nh should be correctly
1671 set by sender, so that the second statement is
1672 just protection against buggy protocols.
1674 skb_reset_mac_header(skb2
);
1676 if (skb_network_header(skb2
) < skb2
->data
||
1677 skb2
->network_header
> skb2
->tail
) {
1678 net_crit_ratelimited("protocol %04x is buggy, dev %s\n",
1679 ntohs(skb2
->protocol
),
1681 skb_reset_network_header(skb2
);
1684 skb2
->transport_header
= skb2
->network_header
;
1685 skb2
->pkt_type
= PACKET_OUTGOING
;
1690 pt_prev
->func(skb2
, skb
->dev
, pt_prev
, skb
->dev
);
1694 /* netif_setup_tc - Handle tc mappings on real_num_tx_queues change
1695 * @dev: Network device
1696 * @txq: number of queues available
1698 * If real_num_tx_queues is changed the tc mappings may no longer be
1699 * valid. To resolve this verify the tc mapping remains valid and if
1700 * not NULL the mapping. With no priorities mapping to this
1701 * offset/count pair it will no longer be used. In the worst case TC0
1702 * is invalid nothing can be done so disable priority mappings. If is
1703 * expected that drivers will fix this mapping if they can before
1704 * calling netif_set_real_num_tx_queues.
1706 static void netif_setup_tc(struct net_device
*dev
, unsigned int txq
)
1709 struct netdev_tc_txq
*tc
= &dev
->tc_to_txq
[0];
1711 /* If TC0 is invalidated disable TC mapping */
1712 if (tc
->offset
+ tc
->count
> txq
) {
1713 pr_warn("Number of in use tx queues changed invalidating tc mappings. Priority traffic classification disabled!\n");
1718 /* Invalidated prio to tc mappings set to TC0 */
1719 for (i
= 1; i
< TC_BITMASK
+ 1; i
++) {
1720 int q
= netdev_get_prio_tc_map(dev
, i
);
1722 tc
= &dev
->tc_to_txq
[q
];
1723 if (tc
->offset
+ tc
->count
> txq
) {
1724 pr_warn("Number of in use tx queues changed. Priority %i to tc mapping %i is no longer valid. Setting map to 0\n",
1726 netdev_set_prio_tc_map(dev
, i
, 0);
1732 * Routine to help set real_num_tx_queues. To avoid skbs mapped to queues
1733 * greater then real_num_tx_queues stale skbs on the qdisc must be flushed.
1735 int netif_set_real_num_tx_queues(struct net_device
*dev
, unsigned int txq
)
1739 if (txq
< 1 || txq
> dev
->num_tx_queues
)
1742 if (dev
->reg_state
== NETREG_REGISTERED
||
1743 dev
->reg_state
== NETREG_UNREGISTERING
) {
1746 rc
= netdev_queue_update_kobjects(dev
, dev
->real_num_tx_queues
,
1752 netif_setup_tc(dev
, txq
);
1754 if (txq
< dev
->real_num_tx_queues
)
1755 qdisc_reset_all_tx_gt(dev
, txq
);
1758 dev
->real_num_tx_queues
= txq
;
1761 EXPORT_SYMBOL(netif_set_real_num_tx_queues
);
1765 * netif_set_real_num_rx_queues - set actual number of RX queues used
1766 * @dev: Network device
1767 * @rxq: Actual number of RX queues
1769 * This must be called either with the rtnl_lock held or before
1770 * registration of the net device. Returns 0 on success, or a
1771 * negative error code. If called before registration, it always
1774 int netif_set_real_num_rx_queues(struct net_device
*dev
, unsigned int rxq
)
1778 if (rxq
< 1 || rxq
> dev
->num_rx_queues
)
1781 if (dev
->reg_state
== NETREG_REGISTERED
) {
1784 rc
= net_rx_queue_update_kobjects(dev
, dev
->real_num_rx_queues
,
1790 dev
->real_num_rx_queues
= rxq
;
1793 EXPORT_SYMBOL(netif_set_real_num_rx_queues
);
1796 static inline void __netif_reschedule(struct Qdisc
*q
)
1798 struct softnet_data
*sd
;
1799 unsigned long flags
;
1801 local_irq_save(flags
);
1802 sd
= &__get_cpu_var(softnet_data
);
1803 q
->next_sched
= NULL
;
1804 *sd
->output_queue_tailp
= q
;
1805 sd
->output_queue_tailp
= &q
->next_sched
;
1806 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1807 local_irq_restore(flags
);
1810 void __netif_schedule(struct Qdisc
*q
)
1812 if (!test_and_set_bit(__QDISC_STATE_SCHED
, &q
->state
))
1813 __netif_reschedule(q
);
1815 EXPORT_SYMBOL(__netif_schedule
);
1817 void dev_kfree_skb_irq(struct sk_buff
*skb
)
1819 if (atomic_dec_and_test(&skb
->users
)) {
1820 struct softnet_data
*sd
;
1821 unsigned long flags
;
1823 local_irq_save(flags
);
1824 sd
= &__get_cpu_var(softnet_data
);
1825 skb
->next
= sd
->completion_queue
;
1826 sd
->completion_queue
= skb
;
1827 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
1828 local_irq_restore(flags
);
1831 EXPORT_SYMBOL(dev_kfree_skb_irq
);
1833 void dev_kfree_skb_any(struct sk_buff
*skb
)
1835 if (in_irq() || irqs_disabled())
1836 dev_kfree_skb_irq(skb
);
1840 EXPORT_SYMBOL(dev_kfree_skb_any
);
1844 * netif_device_detach - mark device as removed
1845 * @dev: network device
1847 * Mark device as removed from system and therefore no longer available.
1849 void netif_device_detach(struct net_device
*dev
)
1851 if (test_and_clear_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1852 netif_running(dev
)) {
1853 netif_tx_stop_all_queues(dev
);
1856 EXPORT_SYMBOL(netif_device_detach
);
1859 * netif_device_attach - mark device as attached
1860 * @dev: network device
1862 * Mark device as attached from system and restart if needed.
1864 void netif_device_attach(struct net_device
*dev
)
1866 if (!test_and_set_bit(__LINK_STATE_PRESENT
, &dev
->state
) &&
1867 netif_running(dev
)) {
1868 netif_tx_wake_all_queues(dev
);
1869 __netdev_watchdog_up(dev
);
1872 EXPORT_SYMBOL(netif_device_attach
);
1874 static void skb_warn_bad_offload(const struct sk_buff
*skb
)
1876 static const netdev_features_t null_features
= 0;
1877 struct net_device
*dev
= skb
->dev
;
1878 const char *driver
= "";
1880 if (dev
&& dev
->dev
.parent
)
1881 driver
= dev_driver_string(dev
->dev
.parent
);
1883 WARN(1, "%s: caps=(%pNF, %pNF) len=%d data_len=%d gso_size=%d "
1884 "gso_type=%d ip_summed=%d\n",
1885 driver
, dev
? &dev
->features
: &null_features
,
1886 skb
->sk
? &skb
->sk
->sk_route_caps
: &null_features
,
1887 skb
->len
, skb
->data_len
, skb_shinfo(skb
)->gso_size
,
1888 skb_shinfo(skb
)->gso_type
, skb
->ip_summed
);
1892 * Invalidate hardware checksum when packet is to be mangled, and
1893 * complete checksum manually on outgoing path.
1895 int skb_checksum_help(struct sk_buff
*skb
)
1898 int ret
= 0, offset
;
1900 if (skb
->ip_summed
== CHECKSUM_COMPLETE
)
1901 goto out_set_summed
;
1903 if (unlikely(skb_shinfo(skb
)->gso_size
)) {
1904 skb_warn_bad_offload(skb
);
1908 offset
= skb_checksum_start_offset(skb
);
1909 BUG_ON(offset
>= skb_headlen(skb
));
1910 csum
= skb_checksum(skb
, offset
, skb
->len
- offset
, 0);
1912 offset
+= skb
->csum_offset
;
1913 BUG_ON(offset
+ sizeof(__sum16
) > skb_headlen(skb
));
1915 if (skb_cloned(skb
) &&
1916 !skb_clone_writable(skb
, offset
+ sizeof(__sum16
))) {
1917 ret
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
);
1922 *(__sum16
*)(skb
->data
+ offset
) = csum_fold(csum
);
1924 skb
->ip_summed
= CHECKSUM_NONE
;
1928 EXPORT_SYMBOL(skb_checksum_help
);
1931 * skb_gso_segment - Perform segmentation on skb.
1932 * @skb: buffer to segment
1933 * @features: features for the output path (see dev->features)
1935 * This function segments the given skb and returns a list of segments.
1937 * It may return NULL if the skb requires no segmentation. This is
1938 * only possible when GSO is used for verifying header integrity.
1940 struct sk_buff
*skb_gso_segment(struct sk_buff
*skb
,
1941 netdev_features_t features
)
1943 struct sk_buff
*segs
= ERR_PTR(-EPROTONOSUPPORT
);
1944 struct packet_type
*ptype
;
1945 __be16 type
= skb
->protocol
;
1946 int vlan_depth
= ETH_HLEN
;
1949 while (type
== htons(ETH_P_8021Q
)) {
1950 struct vlan_hdr
*vh
;
1952 if (unlikely(!pskb_may_pull(skb
, vlan_depth
+ VLAN_HLEN
)))
1953 return ERR_PTR(-EINVAL
);
1955 vh
= (struct vlan_hdr
*)(skb
->data
+ vlan_depth
);
1956 type
= vh
->h_vlan_encapsulated_proto
;
1957 vlan_depth
+= VLAN_HLEN
;
1960 skb_reset_mac_header(skb
);
1961 skb
->mac_len
= skb
->network_header
- skb
->mac_header
;
1962 __skb_pull(skb
, skb
->mac_len
);
1964 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1965 skb_warn_bad_offload(skb
);
1967 if (skb_header_cloned(skb
) &&
1968 (err
= pskb_expand_head(skb
, 0, 0, GFP_ATOMIC
)))
1969 return ERR_PTR(err
);
1973 list_for_each_entry_rcu(ptype
,
1974 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
1975 if (ptype
->type
== type
&& !ptype
->dev
&& ptype
->gso_segment
) {
1976 if (unlikely(skb
->ip_summed
!= CHECKSUM_PARTIAL
)) {
1977 err
= ptype
->gso_send_check(skb
);
1978 segs
= ERR_PTR(err
);
1979 if (err
|| skb_gso_ok(skb
, features
))
1981 __skb_push(skb
, (skb
->data
-
1982 skb_network_header(skb
)));
1984 segs
= ptype
->gso_segment(skb
, features
);
1990 __skb_push(skb
, skb
->data
- skb_mac_header(skb
));
1994 EXPORT_SYMBOL(skb_gso_segment
);
1996 /* Take action when hardware reception checksum errors are detected. */
1998 void netdev_rx_csum_fault(struct net_device
*dev
)
2000 if (net_ratelimit()) {
2001 pr_err("%s: hw csum failure\n", dev
? dev
->name
: "<unknown>");
2005 EXPORT_SYMBOL(netdev_rx_csum_fault
);
2008 /* Actually, we should eliminate this check as soon as we know, that:
2009 * 1. IOMMU is present and allows to map all the memory.
2010 * 2. No high memory really exists on this machine.
2013 static int illegal_highdma(struct net_device
*dev
, struct sk_buff
*skb
)
2015 #ifdef CONFIG_HIGHMEM
2017 if (!(dev
->features
& NETIF_F_HIGHDMA
)) {
2018 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2019 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2020 if (PageHighMem(skb_frag_page(frag
)))
2025 if (PCI_DMA_BUS_IS_PHYS
) {
2026 struct device
*pdev
= dev
->dev
.parent
;
2030 for (i
= 0; i
< skb_shinfo(skb
)->nr_frags
; i
++) {
2031 skb_frag_t
*frag
= &skb_shinfo(skb
)->frags
[i
];
2032 dma_addr_t addr
= page_to_phys(skb_frag_page(frag
));
2033 if (!pdev
->dma_mask
|| addr
+ PAGE_SIZE
- 1 > *pdev
->dma_mask
)
2042 void (*destructor
)(struct sk_buff
*skb
);
2045 #define DEV_GSO_CB(skb) ((struct dev_gso_cb *)(skb)->cb)
2047 static void dev_gso_skb_destructor(struct sk_buff
*skb
)
2049 struct dev_gso_cb
*cb
;
2052 struct sk_buff
*nskb
= skb
->next
;
2054 skb
->next
= nskb
->next
;
2057 } while (skb
->next
);
2059 cb
= DEV_GSO_CB(skb
);
2061 cb
->destructor(skb
);
2065 * dev_gso_segment - Perform emulated hardware segmentation on skb.
2066 * @skb: buffer to segment
2067 * @features: device features as applicable to this skb
2069 * This function segments the given skb and stores the list of segments
2072 static int dev_gso_segment(struct sk_buff
*skb
, netdev_features_t features
)
2074 struct sk_buff
*segs
;
2076 segs
= skb_gso_segment(skb
, features
);
2078 /* Verifying header integrity only. */
2083 return PTR_ERR(segs
);
2086 DEV_GSO_CB(skb
)->destructor
= skb
->destructor
;
2087 skb
->destructor
= dev_gso_skb_destructor
;
2092 static bool can_checksum_protocol(netdev_features_t features
, __be16 protocol
)
2094 return ((features
& NETIF_F_GEN_CSUM
) ||
2095 ((features
& NETIF_F_V4_CSUM
) &&
2096 protocol
== htons(ETH_P_IP
)) ||
2097 ((features
& NETIF_F_V6_CSUM
) &&
2098 protocol
== htons(ETH_P_IPV6
)) ||
2099 ((features
& NETIF_F_FCOE_CRC
) &&
2100 protocol
== htons(ETH_P_FCOE
)));
2103 static netdev_features_t
harmonize_features(struct sk_buff
*skb
,
2104 __be16 protocol
, netdev_features_t features
)
2106 if (!can_checksum_protocol(features
, protocol
)) {
2107 features
&= ~NETIF_F_ALL_CSUM
;
2108 features
&= ~NETIF_F_SG
;
2109 } else if (illegal_highdma(skb
->dev
, skb
)) {
2110 features
&= ~NETIF_F_SG
;
2116 netdev_features_t
netif_skb_features(struct sk_buff
*skb
)
2118 __be16 protocol
= skb
->protocol
;
2119 netdev_features_t features
= skb
->dev
->features
;
2121 if (protocol
== htons(ETH_P_8021Q
)) {
2122 struct vlan_ethhdr
*veh
= (struct vlan_ethhdr
*)skb
->data
;
2123 protocol
= veh
->h_vlan_encapsulated_proto
;
2124 } else if (!vlan_tx_tag_present(skb
)) {
2125 return harmonize_features(skb
, protocol
, features
);
2128 features
&= (skb
->dev
->vlan_features
| NETIF_F_HW_VLAN_TX
);
2130 if (protocol
!= htons(ETH_P_8021Q
)) {
2131 return harmonize_features(skb
, protocol
, features
);
2133 features
&= NETIF_F_SG
| NETIF_F_HIGHDMA
| NETIF_F_FRAGLIST
|
2134 NETIF_F_GEN_CSUM
| NETIF_F_HW_VLAN_TX
;
2135 return harmonize_features(skb
, protocol
, features
);
2138 EXPORT_SYMBOL(netif_skb_features
);
2141 * Returns true if either:
2142 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2143 * 2. skb is fragmented and the device does not support SG, or if
2144 * at least one of fragments is in highmem and device does not
2145 * support DMA from it.
2147 static inline int skb_needs_linearize(struct sk_buff
*skb
,
2150 return skb_is_nonlinear(skb
) &&
2151 ((skb_has_frag_list(skb
) &&
2152 !(features
& NETIF_F_FRAGLIST
)) ||
2153 (skb_shinfo(skb
)->nr_frags
&&
2154 !(features
& NETIF_F_SG
)));
2157 int dev_hard_start_xmit(struct sk_buff
*skb
, struct net_device
*dev
,
2158 struct netdev_queue
*txq
)
2160 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2161 int rc
= NETDEV_TX_OK
;
2162 unsigned int skb_len
;
2164 if (likely(!skb
->next
)) {
2165 netdev_features_t features
;
2168 * If device doesn't need skb->dst, release it right now while
2169 * its hot in this cpu cache
2171 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2174 if (!list_empty(&ptype_all
))
2175 dev_queue_xmit_nit(skb
, dev
);
2177 features
= netif_skb_features(skb
);
2179 if (vlan_tx_tag_present(skb
) &&
2180 !(features
& NETIF_F_HW_VLAN_TX
)) {
2181 skb
= __vlan_put_tag(skb
, vlan_tx_tag_get(skb
));
2188 if (netif_needs_gso(skb
, features
)) {
2189 if (unlikely(dev_gso_segment(skb
, features
)))
2194 if (skb_needs_linearize(skb
, features
) &&
2195 __skb_linearize(skb
))
2198 /* If packet is not checksummed and device does not
2199 * support checksumming for this protocol, complete
2200 * checksumming here.
2202 if (skb
->ip_summed
== CHECKSUM_PARTIAL
) {
2203 skb_set_transport_header(skb
,
2204 skb_checksum_start_offset(skb
));
2205 if (!(features
& NETIF_F_ALL_CSUM
) &&
2206 skb_checksum_help(skb
))
2212 rc
= ops
->ndo_start_xmit(skb
, dev
);
2213 trace_net_dev_xmit(skb
, rc
, dev
, skb_len
);
2214 if (rc
== NETDEV_TX_OK
)
2215 txq_trans_update(txq
);
2221 struct sk_buff
*nskb
= skb
->next
;
2223 skb
->next
= nskb
->next
;
2227 * If device doesn't need nskb->dst, release it right now while
2228 * its hot in this cpu cache
2230 if (dev
->priv_flags
& IFF_XMIT_DST_RELEASE
)
2233 skb_len
= nskb
->len
;
2234 rc
= ops
->ndo_start_xmit(nskb
, dev
);
2235 trace_net_dev_xmit(nskb
, rc
, dev
, skb_len
);
2236 if (unlikely(rc
!= NETDEV_TX_OK
)) {
2237 if (rc
& ~NETDEV_TX_MASK
)
2238 goto out_kfree_gso_skb
;
2239 nskb
->next
= skb
->next
;
2243 txq_trans_update(txq
);
2244 if (unlikely(netif_xmit_stopped(txq
) && skb
->next
))
2245 return NETDEV_TX_BUSY
;
2246 } while (skb
->next
);
2249 if (likely(skb
->next
== NULL
))
2250 skb
->destructor
= DEV_GSO_CB(skb
)->destructor
;
2257 static u32 hashrnd __read_mostly
;
2260 * Returns a Tx hash based on the given packet descriptor a Tx queues' number
2261 * to be used as a distribution range.
2263 u16
__skb_tx_hash(const struct net_device
*dev
, const struct sk_buff
*skb
,
2264 unsigned int num_tx_queues
)
2268 u16 qcount
= num_tx_queues
;
2270 if (skb_rx_queue_recorded(skb
)) {
2271 hash
= skb_get_rx_queue(skb
);
2272 while (unlikely(hash
>= num_tx_queues
))
2273 hash
-= num_tx_queues
;
2278 u8 tc
= netdev_get_prio_tc_map(dev
, skb
->priority
);
2279 qoffset
= dev
->tc_to_txq
[tc
].offset
;
2280 qcount
= dev
->tc_to_txq
[tc
].count
;
2283 if (skb
->sk
&& skb
->sk
->sk_hash
)
2284 hash
= skb
->sk
->sk_hash
;
2286 hash
= (__force u16
) skb
->protocol
;
2287 hash
= jhash_1word(hash
, hashrnd
);
2289 return (u16
) (((u64
) hash
* qcount
) >> 32) + qoffset
;
2291 EXPORT_SYMBOL(__skb_tx_hash
);
2293 static inline u16
dev_cap_txqueue(struct net_device
*dev
, u16 queue_index
)
2295 if (unlikely(queue_index
>= dev
->real_num_tx_queues
)) {
2296 net_warn_ratelimited("%s selects TX queue %d, but real number of TX queues is %d\n",
2297 dev
->name
, queue_index
,
2298 dev
->real_num_tx_queues
);
2304 static inline int get_xps_queue(struct net_device
*dev
, struct sk_buff
*skb
)
2307 struct xps_dev_maps
*dev_maps
;
2308 struct xps_map
*map
;
2309 int queue_index
= -1;
2312 dev_maps
= rcu_dereference(dev
->xps_maps
);
2314 map
= rcu_dereference(
2315 dev_maps
->cpu_map
[raw_smp_processor_id()]);
2318 queue_index
= map
->queues
[0];
2321 if (skb
->sk
&& skb
->sk
->sk_hash
)
2322 hash
= skb
->sk
->sk_hash
;
2324 hash
= (__force u16
) skb
->protocol
^
2326 hash
= jhash_1word(hash
, hashrnd
);
2327 queue_index
= map
->queues
[
2328 ((u64
)hash
* map
->len
) >> 32];
2330 if (unlikely(queue_index
>= dev
->real_num_tx_queues
))
2342 static struct netdev_queue
*dev_pick_tx(struct net_device
*dev
,
2343 struct sk_buff
*skb
)
2346 const struct net_device_ops
*ops
= dev
->netdev_ops
;
2348 if (dev
->real_num_tx_queues
== 1)
2350 else if (ops
->ndo_select_queue
) {
2351 queue_index
= ops
->ndo_select_queue(dev
, skb
);
2352 queue_index
= dev_cap_txqueue(dev
, queue_index
);
2354 struct sock
*sk
= skb
->sk
;
2355 queue_index
= sk_tx_queue_get(sk
);
2357 if (queue_index
< 0 || skb
->ooo_okay
||
2358 queue_index
>= dev
->real_num_tx_queues
) {
2359 int old_index
= queue_index
;
2361 queue_index
= get_xps_queue(dev
, skb
);
2362 if (queue_index
< 0)
2363 queue_index
= skb_tx_hash(dev
, skb
);
2365 if (queue_index
!= old_index
&& sk
) {
2366 struct dst_entry
*dst
=
2367 rcu_dereference_check(sk
->sk_dst_cache
, 1);
2369 if (dst
&& skb_dst(skb
) == dst
)
2370 sk_tx_queue_set(sk
, queue_index
);
2375 skb_set_queue_mapping(skb
, queue_index
);
2376 return netdev_get_tx_queue(dev
, queue_index
);
2379 static inline int __dev_xmit_skb(struct sk_buff
*skb
, struct Qdisc
*q
,
2380 struct net_device
*dev
,
2381 struct netdev_queue
*txq
)
2383 spinlock_t
*root_lock
= qdisc_lock(q
);
2387 qdisc_skb_cb(skb
)->pkt_len
= skb
->len
;
2388 qdisc_calculate_pkt_len(skb
, q
);
2390 * Heuristic to force contended enqueues to serialize on a
2391 * separate lock before trying to get qdisc main lock.
2392 * This permits __QDISC_STATE_RUNNING owner to get the lock more often
2393 * and dequeue packets faster.
2395 contended
= qdisc_is_running(q
);
2396 if (unlikely(contended
))
2397 spin_lock(&q
->busylock
);
2399 spin_lock(root_lock
);
2400 if (unlikely(test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
))) {
2403 } else if ((q
->flags
& TCQ_F_CAN_BYPASS
) && !qdisc_qlen(q
) &&
2404 qdisc_run_begin(q
)) {
2406 * This is a work-conserving queue; there are no old skbs
2407 * waiting to be sent out; and the qdisc is not running -
2408 * xmit the skb directly.
2410 if (!(dev
->priv_flags
& IFF_XMIT_DST_RELEASE
))
2413 qdisc_bstats_update(q
, skb
);
2415 if (sch_direct_xmit(skb
, q
, dev
, txq
, root_lock
)) {
2416 if (unlikely(contended
)) {
2417 spin_unlock(&q
->busylock
);
2424 rc
= NET_XMIT_SUCCESS
;
2427 rc
= q
->enqueue(skb
, q
) & NET_XMIT_MASK
;
2428 if (qdisc_run_begin(q
)) {
2429 if (unlikely(contended
)) {
2430 spin_unlock(&q
->busylock
);
2436 spin_unlock(root_lock
);
2437 if (unlikely(contended
))
2438 spin_unlock(&q
->busylock
);
2442 #if IS_ENABLED(CONFIG_NETPRIO_CGROUP)
2443 static void skb_update_prio(struct sk_buff
*skb
)
2445 struct netprio_map
*map
= rcu_dereference_bh(skb
->dev
->priomap
);
2447 if (!skb
->priority
&& skb
->sk
&& map
) {
2448 unsigned int prioidx
= skb
->sk
->sk_cgrp_prioidx
;
2450 if (prioidx
< map
->priomap_len
)
2451 skb
->priority
= map
->priomap
[prioidx
];
2455 #define skb_update_prio(skb)
2458 static DEFINE_PER_CPU(int, xmit_recursion
);
2459 #define RECURSION_LIMIT 10
2462 * dev_queue_xmit - transmit a buffer
2463 * @skb: buffer to transmit
2465 * Queue a buffer for transmission to a network device. The caller must
2466 * have set the device and priority and built the buffer before calling
2467 * this function. The function can be called from an interrupt.
2469 * A negative errno code is returned on a failure. A success does not
2470 * guarantee the frame will be transmitted as it may be dropped due
2471 * to congestion or traffic shaping.
2473 * -----------------------------------------------------------------------------------
2474 * I notice this method can also return errors from the queue disciplines,
2475 * including NET_XMIT_DROP, which is a positive value. So, errors can also
2478 * Regardless of the return value, the skb is consumed, so it is currently
2479 * difficult to retry a send to this method. (You can bump the ref count
2480 * before sending to hold a reference for retry if you are careful.)
2482 * When calling this method, interrupts MUST be enabled. This is because
2483 * the BH enable code must have IRQs enabled so that it will not deadlock.
2486 int dev_queue_xmit(struct sk_buff
*skb
)
2488 struct net_device
*dev
= skb
->dev
;
2489 struct netdev_queue
*txq
;
2493 /* Disable soft irqs for various locks below. Also
2494 * stops preemption for RCU.
2498 skb_update_prio(skb
);
2500 txq
= dev_pick_tx(dev
, skb
);
2501 q
= rcu_dereference_bh(txq
->qdisc
);
2503 #ifdef CONFIG_NET_CLS_ACT
2504 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_EGRESS
);
2506 trace_net_dev_queue(skb
);
2508 rc
= __dev_xmit_skb(skb
, q
, dev
, txq
);
2512 /* The device has no queue. Common case for software devices:
2513 loopback, all the sorts of tunnels...
2515 Really, it is unlikely that netif_tx_lock protection is necessary
2516 here. (f.e. loopback and IP tunnels are clean ignoring statistics
2518 However, it is possible, that they rely on protection
2521 Check this and shot the lock. It is not prone from deadlocks.
2522 Either shot noqueue qdisc, it is even simpler 8)
2524 if (dev
->flags
& IFF_UP
) {
2525 int cpu
= smp_processor_id(); /* ok because BHs are off */
2527 if (txq
->xmit_lock_owner
!= cpu
) {
2529 if (__this_cpu_read(xmit_recursion
) > RECURSION_LIMIT
)
2530 goto recursion_alert
;
2532 HARD_TX_LOCK(dev
, txq
, cpu
);
2534 if (!netif_xmit_stopped(txq
)) {
2535 __this_cpu_inc(xmit_recursion
);
2536 rc
= dev_hard_start_xmit(skb
, dev
, txq
);
2537 __this_cpu_dec(xmit_recursion
);
2538 if (dev_xmit_complete(rc
)) {
2539 HARD_TX_UNLOCK(dev
, txq
);
2543 HARD_TX_UNLOCK(dev
, txq
);
2544 net_crit_ratelimited("Virtual device %s asks to queue packet!\n",
2547 /* Recursion is detected! It is possible,
2551 net_crit_ratelimited("Dead loop on virtual device %s, fix it urgently!\n",
2557 rcu_read_unlock_bh();
2562 rcu_read_unlock_bh();
2565 EXPORT_SYMBOL(dev_queue_xmit
);
2568 /*=======================================================================
2570 =======================================================================*/
2572 int netdev_max_backlog __read_mostly
= 1000;
2573 int netdev_tstamp_prequeue __read_mostly
= 1;
2574 int netdev_budget __read_mostly
= 300;
2575 int weight_p __read_mostly
= 64; /* old backlog weight */
2577 /* Called with irq disabled */
2578 static inline void ____napi_schedule(struct softnet_data
*sd
,
2579 struct napi_struct
*napi
)
2581 list_add_tail(&napi
->poll_list
, &sd
->poll_list
);
2582 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2586 * __skb_get_rxhash: calculate a flow hash based on src/dst addresses
2587 * and src/dst port numbers. Sets rxhash in skb to non-zero hash value
2588 * on success, zero indicates no valid hash. Also, sets l4_rxhash in skb
2589 * if hash is a canonical 4-tuple hash over transport ports.
2591 void __skb_get_rxhash(struct sk_buff
*skb
)
2593 struct flow_keys keys
;
2596 if (!skb_flow_dissect(skb
, &keys
))
2600 if ((__force u16
)keys
.port16
[1] < (__force u16
)keys
.port16
[0])
2601 swap(keys
.port16
[0], keys
.port16
[1]);
2605 /* get a consistent hash (same value on both flow directions) */
2606 if ((__force u32
)keys
.dst
< (__force u32
)keys
.src
)
2607 swap(keys
.dst
, keys
.src
);
2609 hash
= jhash_3words((__force u32
)keys
.dst
,
2610 (__force u32
)keys
.src
,
2611 (__force u32
)keys
.ports
, hashrnd
);
2617 EXPORT_SYMBOL(__skb_get_rxhash
);
2621 /* One global table that all flow-based protocols share. */
2622 struct rps_sock_flow_table __rcu
*rps_sock_flow_table __read_mostly
;
2623 EXPORT_SYMBOL(rps_sock_flow_table
);
2625 struct static_key rps_needed __read_mostly
;
2627 static struct rps_dev_flow
*
2628 set_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2629 struct rps_dev_flow
*rflow
, u16 next_cpu
)
2631 if (next_cpu
!= RPS_NO_CPU
) {
2632 #ifdef CONFIG_RFS_ACCEL
2633 struct netdev_rx_queue
*rxqueue
;
2634 struct rps_dev_flow_table
*flow_table
;
2635 struct rps_dev_flow
*old_rflow
;
2640 /* Should we steer this flow to a different hardware queue? */
2641 if (!skb_rx_queue_recorded(skb
) || !dev
->rx_cpu_rmap
||
2642 !(dev
->features
& NETIF_F_NTUPLE
))
2644 rxq_index
= cpu_rmap_lookup_index(dev
->rx_cpu_rmap
, next_cpu
);
2645 if (rxq_index
== skb_get_rx_queue(skb
))
2648 rxqueue
= dev
->_rx
+ rxq_index
;
2649 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2652 flow_id
= skb
->rxhash
& flow_table
->mask
;
2653 rc
= dev
->netdev_ops
->ndo_rx_flow_steer(dev
, skb
,
2654 rxq_index
, flow_id
);
2658 rflow
= &flow_table
->flows
[flow_id
];
2660 if (old_rflow
->filter
== rflow
->filter
)
2661 old_rflow
->filter
= RPS_NO_FILTER
;
2665 per_cpu(softnet_data
, next_cpu
).input_queue_head
;
2668 rflow
->cpu
= next_cpu
;
2673 * get_rps_cpu is called from netif_receive_skb and returns the target
2674 * CPU from the RPS map of the receiving queue for a given skb.
2675 * rcu_read_lock must be held on entry.
2677 static int get_rps_cpu(struct net_device
*dev
, struct sk_buff
*skb
,
2678 struct rps_dev_flow
**rflowp
)
2680 struct netdev_rx_queue
*rxqueue
;
2681 struct rps_map
*map
;
2682 struct rps_dev_flow_table
*flow_table
;
2683 struct rps_sock_flow_table
*sock_flow_table
;
2687 if (skb_rx_queue_recorded(skb
)) {
2688 u16 index
= skb_get_rx_queue(skb
);
2689 if (unlikely(index
>= dev
->real_num_rx_queues
)) {
2690 WARN_ONCE(dev
->real_num_rx_queues
> 1,
2691 "%s received packet on queue %u, but number "
2692 "of RX queues is %u\n",
2693 dev
->name
, index
, dev
->real_num_rx_queues
);
2696 rxqueue
= dev
->_rx
+ index
;
2700 map
= rcu_dereference(rxqueue
->rps_map
);
2702 if (map
->len
== 1 &&
2703 !rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2704 tcpu
= map
->cpus
[0];
2705 if (cpu_online(tcpu
))
2709 } else if (!rcu_access_pointer(rxqueue
->rps_flow_table
)) {
2713 skb_reset_network_header(skb
);
2714 if (!skb_get_rxhash(skb
))
2717 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2718 sock_flow_table
= rcu_dereference(rps_sock_flow_table
);
2719 if (flow_table
&& sock_flow_table
) {
2721 struct rps_dev_flow
*rflow
;
2723 rflow
= &flow_table
->flows
[skb
->rxhash
& flow_table
->mask
];
2726 next_cpu
= sock_flow_table
->ents
[skb
->rxhash
&
2727 sock_flow_table
->mask
];
2730 * If the desired CPU (where last recvmsg was done) is
2731 * different from current CPU (one in the rx-queue flow
2732 * table entry), switch if one of the following holds:
2733 * - Current CPU is unset (equal to RPS_NO_CPU).
2734 * - Current CPU is offline.
2735 * - The current CPU's queue tail has advanced beyond the
2736 * last packet that was enqueued using this table entry.
2737 * This guarantees that all previous packets for the flow
2738 * have been dequeued, thus preserving in order delivery.
2740 if (unlikely(tcpu
!= next_cpu
) &&
2741 (tcpu
== RPS_NO_CPU
|| !cpu_online(tcpu
) ||
2742 ((int)(per_cpu(softnet_data
, tcpu
).input_queue_head
-
2743 rflow
->last_qtail
)) >= 0))
2744 rflow
= set_rps_cpu(dev
, skb
, rflow
, next_cpu
);
2746 if (tcpu
!= RPS_NO_CPU
&& cpu_online(tcpu
)) {
2754 tcpu
= map
->cpus
[((u64
) skb
->rxhash
* map
->len
) >> 32];
2756 if (cpu_online(tcpu
)) {
2766 #ifdef CONFIG_RFS_ACCEL
2769 * rps_may_expire_flow - check whether an RFS hardware filter may be removed
2770 * @dev: Device on which the filter was set
2771 * @rxq_index: RX queue index
2772 * @flow_id: Flow ID passed to ndo_rx_flow_steer()
2773 * @filter_id: Filter ID returned by ndo_rx_flow_steer()
2775 * Drivers that implement ndo_rx_flow_steer() should periodically call
2776 * this function for each installed filter and remove the filters for
2777 * which it returns %true.
2779 bool rps_may_expire_flow(struct net_device
*dev
, u16 rxq_index
,
2780 u32 flow_id
, u16 filter_id
)
2782 struct netdev_rx_queue
*rxqueue
= dev
->_rx
+ rxq_index
;
2783 struct rps_dev_flow_table
*flow_table
;
2784 struct rps_dev_flow
*rflow
;
2789 flow_table
= rcu_dereference(rxqueue
->rps_flow_table
);
2790 if (flow_table
&& flow_id
<= flow_table
->mask
) {
2791 rflow
= &flow_table
->flows
[flow_id
];
2792 cpu
= ACCESS_ONCE(rflow
->cpu
);
2793 if (rflow
->filter
== filter_id
&& cpu
!= RPS_NO_CPU
&&
2794 ((int)(per_cpu(softnet_data
, cpu
).input_queue_head
-
2795 rflow
->last_qtail
) <
2796 (int)(10 * flow_table
->mask
)))
2802 EXPORT_SYMBOL(rps_may_expire_flow
);
2804 #endif /* CONFIG_RFS_ACCEL */
2806 /* Called from hardirq (IPI) context */
2807 static void rps_trigger_softirq(void *data
)
2809 struct softnet_data
*sd
= data
;
2811 ____napi_schedule(sd
, &sd
->backlog
);
2815 #endif /* CONFIG_RPS */
2818 * Check if this softnet_data structure is another cpu one
2819 * If yes, queue it to our IPI list and return 1
2822 static int rps_ipi_queued(struct softnet_data
*sd
)
2825 struct softnet_data
*mysd
= &__get_cpu_var(softnet_data
);
2828 sd
->rps_ipi_next
= mysd
->rps_ipi_list
;
2829 mysd
->rps_ipi_list
= sd
;
2831 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
2834 #endif /* CONFIG_RPS */
2839 * enqueue_to_backlog is called to queue an skb to a per CPU backlog
2840 * queue (may be a remote CPU queue).
2842 static int enqueue_to_backlog(struct sk_buff
*skb
, int cpu
,
2843 unsigned int *qtail
)
2845 struct softnet_data
*sd
;
2846 unsigned long flags
;
2848 sd
= &per_cpu(softnet_data
, cpu
);
2850 local_irq_save(flags
);
2853 if (skb_queue_len(&sd
->input_pkt_queue
) <= netdev_max_backlog
) {
2854 if (skb_queue_len(&sd
->input_pkt_queue
)) {
2856 __skb_queue_tail(&sd
->input_pkt_queue
, skb
);
2857 input_queue_tail_incr_save(sd
, qtail
);
2859 local_irq_restore(flags
);
2860 return NET_RX_SUCCESS
;
2863 /* Schedule NAPI for backlog device
2864 * We can use non atomic operation since we own the queue lock
2866 if (!__test_and_set_bit(NAPI_STATE_SCHED
, &sd
->backlog
.state
)) {
2867 if (!rps_ipi_queued(sd
))
2868 ____napi_schedule(sd
, &sd
->backlog
);
2876 local_irq_restore(flags
);
2878 atomic_long_inc(&skb
->dev
->rx_dropped
);
2884 * netif_rx - post buffer to the network code
2885 * @skb: buffer to post
2887 * This function receives a packet from a device driver and queues it for
2888 * the upper (protocol) levels to process. It always succeeds. The buffer
2889 * may be dropped during processing for congestion control or by the
2893 * NET_RX_SUCCESS (no congestion)
2894 * NET_RX_DROP (packet was dropped)
2898 int netif_rx(struct sk_buff
*skb
)
2902 /* if netpoll wants it, pretend we never saw it */
2903 if (netpoll_rx(skb
))
2906 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
2908 trace_netif_rx(skb
);
2910 if (static_key_false(&rps_needed
)) {
2911 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
2917 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
2919 cpu
= smp_processor_id();
2921 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
2929 ret
= enqueue_to_backlog(skb
, get_cpu(), &qtail
);
2934 EXPORT_SYMBOL(netif_rx
);
2936 int netif_rx_ni(struct sk_buff
*skb
)
2941 err
= netif_rx(skb
);
2942 if (local_softirq_pending())
2948 EXPORT_SYMBOL(netif_rx_ni
);
2950 static void net_tx_action(struct softirq_action
*h
)
2952 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
2954 if (sd
->completion_queue
) {
2955 struct sk_buff
*clist
;
2957 local_irq_disable();
2958 clist
= sd
->completion_queue
;
2959 sd
->completion_queue
= NULL
;
2963 struct sk_buff
*skb
= clist
;
2964 clist
= clist
->next
;
2966 WARN_ON(atomic_read(&skb
->users
));
2967 trace_kfree_skb(skb
, net_tx_action
);
2972 if (sd
->output_queue
) {
2975 local_irq_disable();
2976 head
= sd
->output_queue
;
2977 sd
->output_queue
= NULL
;
2978 sd
->output_queue_tailp
= &sd
->output_queue
;
2982 struct Qdisc
*q
= head
;
2983 spinlock_t
*root_lock
;
2985 head
= head
->next_sched
;
2987 root_lock
= qdisc_lock(q
);
2988 if (spin_trylock(root_lock
)) {
2989 smp_mb__before_clear_bit();
2990 clear_bit(__QDISC_STATE_SCHED
,
2993 spin_unlock(root_lock
);
2995 if (!test_bit(__QDISC_STATE_DEACTIVATED
,
2997 __netif_reschedule(q
);
2999 smp_mb__before_clear_bit();
3000 clear_bit(__QDISC_STATE_SCHED
,
3008 #if (defined(CONFIG_BRIDGE) || defined(CONFIG_BRIDGE_MODULE)) && \
3009 (defined(CONFIG_ATM_LANE) || defined(CONFIG_ATM_LANE_MODULE))
3010 /* This hook is defined here for ATM LANE */
3011 int (*br_fdb_test_addr_hook
)(struct net_device
*dev
,
3012 unsigned char *addr
) __read_mostly
;
3013 EXPORT_SYMBOL_GPL(br_fdb_test_addr_hook
);
3016 #ifdef CONFIG_NET_CLS_ACT
3017 /* TODO: Maybe we should just force sch_ingress to be compiled in
3018 * when CONFIG_NET_CLS_ACT is? otherwise some useless instructions
3019 * a compare and 2 stores extra right now if we dont have it on
3020 * but have CONFIG_NET_CLS_ACT
3021 * NOTE: This doesn't stop any functionality; if you dont have
3022 * the ingress scheduler, you just can't add policies on ingress.
3025 static int ing_filter(struct sk_buff
*skb
, struct netdev_queue
*rxq
)
3027 struct net_device
*dev
= skb
->dev
;
3028 u32 ttl
= G_TC_RTTL(skb
->tc_verd
);
3029 int result
= TC_ACT_OK
;
3032 if (unlikely(MAX_RED_LOOP
< ttl
++)) {
3033 net_warn_ratelimited("Redir loop detected Dropping packet (%d->%d)\n",
3034 skb
->skb_iif
, dev
->ifindex
);
3038 skb
->tc_verd
= SET_TC_RTTL(skb
->tc_verd
, ttl
);
3039 skb
->tc_verd
= SET_TC_AT(skb
->tc_verd
, AT_INGRESS
);
3042 if (q
!= &noop_qdisc
) {
3043 spin_lock(qdisc_lock(q
));
3044 if (likely(!test_bit(__QDISC_STATE_DEACTIVATED
, &q
->state
)))
3045 result
= qdisc_enqueue_root(skb
, q
);
3046 spin_unlock(qdisc_lock(q
));
3052 static inline struct sk_buff
*handle_ing(struct sk_buff
*skb
,
3053 struct packet_type
**pt_prev
,
3054 int *ret
, struct net_device
*orig_dev
)
3056 struct netdev_queue
*rxq
= rcu_dereference(skb
->dev
->ingress_queue
);
3058 if (!rxq
|| rxq
->qdisc
== &noop_qdisc
)
3062 *ret
= deliver_skb(skb
, *pt_prev
, orig_dev
);
3066 switch (ing_filter(skb
, rxq
)) {
3080 * netdev_rx_handler_register - register receive handler
3081 * @dev: device to register a handler for
3082 * @rx_handler: receive handler to register
3083 * @rx_handler_data: data pointer that is used by rx handler
3085 * Register a receive hander for a device. This handler will then be
3086 * called from __netif_receive_skb. A negative errno code is returned
3089 * The caller must hold the rtnl_mutex.
3091 * For a general description of rx_handler, see enum rx_handler_result.
3093 int netdev_rx_handler_register(struct net_device
*dev
,
3094 rx_handler_func_t
*rx_handler
,
3095 void *rx_handler_data
)
3099 if (dev
->rx_handler
)
3102 rcu_assign_pointer(dev
->rx_handler_data
, rx_handler_data
);
3103 rcu_assign_pointer(dev
->rx_handler
, rx_handler
);
3107 EXPORT_SYMBOL_GPL(netdev_rx_handler_register
);
3110 * netdev_rx_handler_unregister - unregister receive handler
3111 * @dev: device to unregister a handler from
3113 * Unregister a receive hander from a device.
3115 * The caller must hold the rtnl_mutex.
3117 void netdev_rx_handler_unregister(struct net_device
*dev
)
3121 RCU_INIT_POINTER(dev
->rx_handler
, NULL
);
3122 RCU_INIT_POINTER(dev
->rx_handler_data
, NULL
);
3124 EXPORT_SYMBOL_GPL(netdev_rx_handler_unregister
);
3126 static int __netif_receive_skb(struct sk_buff
*skb
)
3128 struct packet_type
*ptype
, *pt_prev
;
3129 rx_handler_func_t
*rx_handler
;
3130 struct net_device
*orig_dev
;
3131 struct net_device
*null_or_dev
;
3132 bool deliver_exact
= false;
3133 int ret
= NET_RX_DROP
;
3136 net_timestamp_check(!netdev_tstamp_prequeue
, skb
);
3138 trace_netif_receive_skb(skb
);
3140 /* if we've gotten here through NAPI, check netpoll */
3141 if (netpoll_receive_skb(skb
))
3145 skb
->skb_iif
= skb
->dev
->ifindex
;
3146 orig_dev
= skb
->dev
;
3148 skb_reset_network_header(skb
);
3149 skb_reset_transport_header(skb
);
3150 skb_reset_mac_len(skb
);
3158 __this_cpu_inc(softnet_data
.processed
);
3160 if (skb
->protocol
== cpu_to_be16(ETH_P_8021Q
)) {
3161 skb
= vlan_untag(skb
);
3166 #ifdef CONFIG_NET_CLS_ACT
3167 if (skb
->tc_verd
& TC_NCLS
) {
3168 skb
->tc_verd
= CLR_TC_NCLS(skb
->tc_verd
);
3173 list_for_each_entry_rcu(ptype
, &ptype_all
, list
) {
3174 if (!ptype
->dev
|| ptype
->dev
== skb
->dev
) {
3176 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3181 #ifdef CONFIG_NET_CLS_ACT
3182 skb
= handle_ing(skb
, &pt_prev
, &ret
, orig_dev
);
3188 rx_handler
= rcu_dereference(skb
->dev
->rx_handler
);
3189 if (vlan_tx_tag_present(skb
)) {
3191 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3194 if (vlan_do_receive(&skb
, !rx_handler
))
3196 else if (unlikely(!skb
))
3202 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3205 switch (rx_handler(&skb
)) {
3206 case RX_HANDLER_CONSUMED
:
3208 case RX_HANDLER_ANOTHER
:
3210 case RX_HANDLER_EXACT
:
3211 deliver_exact
= true;
3212 case RX_HANDLER_PASS
:
3219 /* deliver only exact match when indicated */
3220 null_or_dev
= deliver_exact
? skb
->dev
: NULL
;
3222 type
= skb
->protocol
;
3223 list_for_each_entry_rcu(ptype
,
3224 &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
], list
) {
3225 if (ptype
->type
== type
&&
3226 (ptype
->dev
== null_or_dev
|| ptype
->dev
== skb
->dev
||
3227 ptype
->dev
== orig_dev
)) {
3229 ret
= deliver_skb(skb
, pt_prev
, orig_dev
);
3235 ret
= pt_prev
->func(skb
, skb
->dev
, pt_prev
, orig_dev
);
3237 atomic_long_inc(&skb
->dev
->rx_dropped
);
3239 /* Jamal, now you will not able to escape explaining
3240 * me how you were going to use this. :-)
3251 * netif_receive_skb - process receive buffer from network
3252 * @skb: buffer to process
3254 * netif_receive_skb() is the main receive data processing function.
3255 * It always succeeds. The buffer may be dropped during processing
3256 * for congestion control or by the protocol layers.
3258 * This function may only be called from softirq context and interrupts
3259 * should be enabled.
3261 * Return values (usually ignored):
3262 * NET_RX_SUCCESS: no congestion
3263 * NET_RX_DROP: packet was dropped
3265 int netif_receive_skb(struct sk_buff
*skb
)
3267 net_timestamp_check(netdev_tstamp_prequeue
, skb
);
3269 if (skb_defer_rx_timestamp(skb
))
3270 return NET_RX_SUCCESS
;
3273 if (static_key_false(&rps_needed
)) {
3274 struct rps_dev_flow voidflow
, *rflow
= &voidflow
;
3279 cpu
= get_rps_cpu(skb
->dev
, skb
, &rflow
);
3282 ret
= enqueue_to_backlog(skb
, cpu
, &rflow
->last_qtail
);
3289 return __netif_receive_skb(skb
);
3291 EXPORT_SYMBOL(netif_receive_skb
);
3293 /* Network device is going away, flush any packets still pending
3294 * Called with irqs disabled.
3296 static void flush_backlog(void *arg
)
3298 struct net_device
*dev
= arg
;
3299 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3300 struct sk_buff
*skb
, *tmp
;
3303 skb_queue_walk_safe(&sd
->input_pkt_queue
, skb
, tmp
) {
3304 if (skb
->dev
== dev
) {
3305 __skb_unlink(skb
, &sd
->input_pkt_queue
);
3307 input_queue_head_incr(sd
);
3312 skb_queue_walk_safe(&sd
->process_queue
, skb
, tmp
) {
3313 if (skb
->dev
== dev
) {
3314 __skb_unlink(skb
, &sd
->process_queue
);
3316 input_queue_head_incr(sd
);
3321 static int napi_gro_complete(struct sk_buff
*skb
)
3323 struct packet_type
*ptype
;
3324 __be16 type
= skb
->protocol
;
3325 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3328 if (NAPI_GRO_CB(skb
)->count
== 1) {
3329 skb_shinfo(skb
)->gso_size
= 0;
3334 list_for_each_entry_rcu(ptype
, head
, list
) {
3335 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_complete
)
3338 err
= ptype
->gro_complete(skb
);
3344 WARN_ON(&ptype
->list
== head
);
3346 return NET_RX_SUCCESS
;
3350 return netif_receive_skb(skb
);
3353 inline void napi_gro_flush(struct napi_struct
*napi
)
3355 struct sk_buff
*skb
, *next
;
3357 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3360 napi_gro_complete(skb
);
3363 napi
->gro_count
= 0;
3364 napi
->gro_list
= NULL
;
3366 EXPORT_SYMBOL(napi_gro_flush
);
3368 enum gro_result
dev_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3370 struct sk_buff
**pp
= NULL
;
3371 struct packet_type
*ptype
;
3372 __be16 type
= skb
->protocol
;
3373 struct list_head
*head
= &ptype_base
[ntohs(type
) & PTYPE_HASH_MASK
];
3376 enum gro_result ret
;
3378 if (!(skb
->dev
->features
& NETIF_F_GRO
) || netpoll_rx_on(skb
))
3381 if (skb_is_gso(skb
) || skb_has_frag_list(skb
))
3385 list_for_each_entry_rcu(ptype
, head
, list
) {
3386 if (ptype
->type
!= type
|| ptype
->dev
|| !ptype
->gro_receive
)
3389 skb_set_network_header(skb
, skb_gro_offset(skb
));
3390 mac_len
= skb
->network_header
- skb
->mac_header
;
3391 skb
->mac_len
= mac_len
;
3392 NAPI_GRO_CB(skb
)->same_flow
= 0;
3393 NAPI_GRO_CB(skb
)->flush
= 0;
3394 NAPI_GRO_CB(skb
)->free
= 0;
3396 pp
= ptype
->gro_receive(&napi
->gro_list
, skb
);
3401 if (&ptype
->list
== head
)
3404 same_flow
= NAPI_GRO_CB(skb
)->same_flow
;
3405 ret
= NAPI_GRO_CB(skb
)->free
? GRO_MERGED_FREE
: GRO_MERGED
;
3408 struct sk_buff
*nskb
= *pp
;
3412 napi_gro_complete(nskb
);
3419 if (NAPI_GRO_CB(skb
)->flush
|| napi
->gro_count
>= MAX_GRO_SKBS
)
3423 NAPI_GRO_CB(skb
)->count
= 1;
3424 skb_shinfo(skb
)->gso_size
= skb_gro_len(skb
);
3425 skb
->next
= napi
->gro_list
;
3426 napi
->gro_list
= skb
;
3430 if (skb_headlen(skb
) < skb_gro_offset(skb
)) {
3431 int grow
= skb_gro_offset(skb
) - skb_headlen(skb
);
3433 BUG_ON(skb
->end
- skb
->tail
< grow
);
3435 memcpy(skb_tail_pointer(skb
), NAPI_GRO_CB(skb
)->frag0
, grow
);
3438 skb
->data_len
-= grow
;
3440 skb_shinfo(skb
)->frags
[0].page_offset
+= grow
;
3441 skb_frag_size_sub(&skb_shinfo(skb
)->frags
[0], grow
);
3443 if (unlikely(!skb_frag_size(&skb_shinfo(skb
)->frags
[0]))) {
3444 skb_frag_unref(skb
, 0);
3445 memmove(skb_shinfo(skb
)->frags
,
3446 skb_shinfo(skb
)->frags
+ 1,
3447 --skb_shinfo(skb
)->nr_frags
* sizeof(skb_frag_t
));
3458 EXPORT_SYMBOL(dev_gro_receive
);
3460 static inline gro_result_t
3461 __napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3464 unsigned int maclen
= skb
->dev
->hard_header_len
;
3466 for (p
= napi
->gro_list
; p
; p
= p
->next
) {
3467 unsigned long diffs
;
3469 diffs
= (unsigned long)p
->dev
^ (unsigned long)skb
->dev
;
3470 diffs
|= p
->vlan_tci
^ skb
->vlan_tci
;
3471 if (maclen
== ETH_HLEN
)
3472 diffs
|= compare_ether_header(skb_mac_header(p
),
3473 skb_gro_mac_header(skb
));
3475 diffs
= memcmp(skb_mac_header(p
),
3476 skb_gro_mac_header(skb
),
3478 NAPI_GRO_CB(p
)->same_flow
= !diffs
;
3479 NAPI_GRO_CB(p
)->flush
= 0;
3482 return dev_gro_receive(napi
, skb
);
3485 gro_result_t
napi_skb_finish(gro_result_t ret
, struct sk_buff
*skb
)
3489 if (netif_receive_skb(skb
))
3497 case GRO_MERGED_FREE
:
3498 if (NAPI_GRO_CB(skb
)->free
== NAPI_GRO_FREE_STOLEN_HEAD
)
3499 kmem_cache_free(skbuff_head_cache
, skb
);
3511 EXPORT_SYMBOL(napi_skb_finish
);
3513 void skb_gro_reset_offset(struct sk_buff
*skb
)
3515 NAPI_GRO_CB(skb
)->data_offset
= 0;
3516 NAPI_GRO_CB(skb
)->frag0
= NULL
;
3517 NAPI_GRO_CB(skb
)->frag0_len
= 0;
3519 if (skb
->mac_header
== skb
->tail
&&
3520 !PageHighMem(skb_frag_page(&skb_shinfo(skb
)->frags
[0]))) {
3521 NAPI_GRO_CB(skb
)->frag0
=
3522 skb_frag_address(&skb_shinfo(skb
)->frags
[0]);
3523 NAPI_GRO_CB(skb
)->frag0_len
= skb_frag_size(&skb_shinfo(skb
)->frags
[0]);
3526 EXPORT_SYMBOL(skb_gro_reset_offset
);
3528 gro_result_t
napi_gro_receive(struct napi_struct
*napi
, struct sk_buff
*skb
)
3530 skb_gro_reset_offset(skb
);
3532 return napi_skb_finish(__napi_gro_receive(napi
, skb
), skb
);
3534 EXPORT_SYMBOL(napi_gro_receive
);
3536 static void napi_reuse_skb(struct napi_struct
*napi
, struct sk_buff
*skb
)
3538 __skb_pull(skb
, skb_headlen(skb
));
3539 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
3540 skb_reserve(skb
, NET_SKB_PAD
+ NET_IP_ALIGN
- skb_headroom(skb
));
3542 skb
->dev
= napi
->dev
;
3548 struct sk_buff
*napi_get_frags(struct napi_struct
*napi
)
3550 struct sk_buff
*skb
= napi
->skb
;
3553 skb
= netdev_alloc_skb_ip_align(napi
->dev
, GRO_MAX_HEAD
);
3559 EXPORT_SYMBOL(napi_get_frags
);
3561 gro_result_t
napi_frags_finish(struct napi_struct
*napi
, struct sk_buff
*skb
,
3567 skb
->protocol
= eth_type_trans(skb
, skb
->dev
);
3569 if (ret
== GRO_HELD
)
3570 skb_gro_pull(skb
, -ETH_HLEN
);
3571 else if (netif_receive_skb(skb
))
3576 case GRO_MERGED_FREE
:
3577 napi_reuse_skb(napi
, skb
);
3586 EXPORT_SYMBOL(napi_frags_finish
);
3588 static struct sk_buff
*napi_frags_skb(struct napi_struct
*napi
)
3590 struct sk_buff
*skb
= napi
->skb
;
3597 skb_reset_mac_header(skb
);
3598 skb_gro_reset_offset(skb
);
3600 off
= skb_gro_offset(skb
);
3601 hlen
= off
+ sizeof(*eth
);
3602 eth
= skb_gro_header_fast(skb
, off
);
3603 if (skb_gro_header_hard(skb
, hlen
)) {
3604 eth
= skb_gro_header_slow(skb
, hlen
, off
);
3605 if (unlikely(!eth
)) {
3606 napi_reuse_skb(napi
, skb
);
3612 skb_gro_pull(skb
, sizeof(*eth
));
3615 * This works because the only protocols we care about don't require
3616 * special handling. We'll fix it up properly at the end.
3618 skb
->protocol
= eth
->h_proto
;
3624 gro_result_t
napi_gro_frags(struct napi_struct
*napi
)
3626 struct sk_buff
*skb
= napi_frags_skb(napi
);
3631 return napi_frags_finish(napi
, skb
, __napi_gro_receive(napi
, skb
));
3633 EXPORT_SYMBOL(napi_gro_frags
);
3636 * net_rps_action sends any pending IPI's for rps.
3637 * Note: called with local irq disabled, but exits with local irq enabled.
3639 static void net_rps_action_and_irq_enable(struct softnet_data
*sd
)
3642 struct softnet_data
*remsd
= sd
->rps_ipi_list
;
3645 sd
->rps_ipi_list
= NULL
;
3649 /* Send pending IPI's to kick RPS processing on remote cpus. */
3651 struct softnet_data
*next
= remsd
->rps_ipi_next
;
3653 if (cpu_online(remsd
->cpu
))
3654 __smp_call_function_single(remsd
->cpu
,
3663 static int process_backlog(struct napi_struct
*napi
, int quota
)
3666 struct softnet_data
*sd
= container_of(napi
, struct softnet_data
, backlog
);
3669 /* Check if we have pending ipi, its better to send them now,
3670 * not waiting net_rx_action() end.
3672 if (sd
->rps_ipi_list
) {
3673 local_irq_disable();
3674 net_rps_action_and_irq_enable(sd
);
3677 napi
->weight
= weight_p
;
3678 local_irq_disable();
3679 while (work
< quota
) {
3680 struct sk_buff
*skb
;
3683 while ((skb
= __skb_dequeue(&sd
->process_queue
))) {
3685 __netif_receive_skb(skb
);
3686 local_irq_disable();
3687 input_queue_head_incr(sd
);
3688 if (++work
>= quota
) {
3695 qlen
= skb_queue_len(&sd
->input_pkt_queue
);
3697 skb_queue_splice_tail_init(&sd
->input_pkt_queue
,
3698 &sd
->process_queue
);
3700 if (qlen
< quota
- work
) {
3702 * Inline a custom version of __napi_complete().
3703 * only current cpu owns and manipulates this napi,
3704 * and NAPI_STATE_SCHED is the only possible flag set on backlog.
3705 * we can use a plain write instead of clear_bit(),
3706 * and we dont need an smp_mb() memory barrier.
3708 list_del(&napi
->poll_list
);
3711 quota
= work
+ qlen
;
3721 * __napi_schedule - schedule for receive
3722 * @n: entry to schedule
3724 * The entry's receive function will be scheduled to run
3726 void __napi_schedule(struct napi_struct
*n
)
3728 unsigned long flags
;
3730 local_irq_save(flags
);
3731 ____napi_schedule(&__get_cpu_var(softnet_data
), n
);
3732 local_irq_restore(flags
);
3734 EXPORT_SYMBOL(__napi_schedule
);
3736 void __napi_complete(struct napi_struct
*n
)
3738 BUG_ON(!test_bit(NAPI_STATE_SCHED
, &n
->state
));
3739 BUG_ON(n
->gro_list
);
3741 list_del(&n
->poll_list
);
3742 smp_mb__before_clear_bit();
3743 clear_bit(NAPI_STATE_SCHED
, &n
->state
);
3745 EXPORT_SYMBOL(__napi_complete
);
3747 void napi_complete(struct napi_struct
*n
)
3749 unsigned long flags
;
3752 * don't let napi dequeue from the cpu poll list
3753 * just in case its running on a different cpu
3755 if (unlikely(test_bit(NAPI_STATE_NPSVC
, &n
->state
)))
3759 local_irq_save(flags
);
3761 local_irq_restore(flags
);
3763 EXPORT_SYMBOL(napi_complete
);
3765 void netif_napi_add(struct net_device
*dev
, struct napi_struct
*napi
,
3766 int (*poll
)(struct napi_struct
*, int), int weight
)
3768 INIT_LIST_HEAD(&napi
->poll_list
);
3769 napi
->gro_count
= 0;
3770 napi
->gro_list
= NULL
;
3773 napi
->weight
= weight
;
3774 list_add(&napi
->dev_list
, &dev
->napi_list
);
3776 #ifdef CONFIG_NETPOLL
3777 spin_lock_init(&napi
->poll_lock
);
3778 napi
->poll_owner
= -1;
3780 set_bit(NAPI_STATE_SCHED
, &napi
->state
);
3782 EXPORT_SYMBOL(netif_napi_add
);
3784 void netif_napi_del(struct napi_struct
*napi
)
3786 struct sk_buff
*skb
, *next
;
3788 list_del_init(&napi
->dev_list
);
3789 napi_free_frags(napi
);
3791 for (skb
= napi
->gro_list
; skb
; skb
= next
) {
3797 napi
->gro_list
= NULL
;
3798 napi
->gro_count
= 0;
3800 EXPORT_SYMBOL(netif_napi_del
);
3802 static void net_rx_action(struct softirq_action
*h
)
3804 struct softnet_data
*sd
= &__get_cpu_var(softnet_data
);
3805 unsigned long time_limit
= jiffies
+ 2;
3806 int budget
= netdev_budget
;
3809 local_irq_disable();
3811 while (!list_empty(&sd
->poll_list
)) {
3812 struct napi_struct
*n
;
3815 /* If softirq window is exhuasted then punt.
3816 * Allow this to run for 2 jiffies since which will allow
3817 * an average latency of 1.5/HZ.
3819 if (unlikely(budget
<= 0 || time_after(jiffies
, time_limit
)))
3824 /* Even though interrupts have been re-enabled, this
3825 * access is safe because interrupts can only add new
3826 * entries to the tail of this list, and only ->poll()
3827 * calls can remove this head entry from the list.
3829 n
= list_first_entry(&sd
->poll_list
, struct napi_struct
, poll_list
);
3831 have
= netpoll_poll_lock(n
);
3835 /* This NAPI_STATE_SCHED test is for avoiding a race
3836 * with netpoll's poll_napi(). Only the entity which
3837 * obtains the lock and sees NAPI_STATE_SCHED set will
3838 * actually make the ->poll() call. Therefore we avoid
3839 * accidentally calling ->poll() when NAPI is not scheduled.
3842 if (test_bit(NAPI_STATE_SCHED
, &n
->state
)) {
3843 work
= n
->poll(n
, weight
);
3847 WARN_ON_ONCE(work
> weight
);
3851 local_irq_disable();
3853 /* Drivers must not modify the NAPI state if they
3854 * consume the entire weight. In such cases this code
3855 * still "owns" the NAPI instance and therefore can
3856 * move the instance around on the list at-will.
3858 if (unlikely(work
== weight
)) {
3859 if (unlikely(napi_disable_pending(n
))) {
3862 local_irq_disable();
3864 list_move_tail(&n
->poll_list
, &sd
->poll_list
);
3867 netpoll_poll_unlock(have
);
3870 net_rps_action_and_irq_enable(sd
);
3872 #ifdef CONFIG_NET_DMA
3874 * There may not be any more sk_buffs coming right now, so push
3875 * any pending DMA copies to hardware
3877 dma_issue_pending_all();
3884 __raise_softirq_irqoff(NET_RX_SOFTIRQ
);
3888 static gifconf_func_t
*gifconf_list
[NPROTO
];
3891 * register_gifconf - register a SIOCGIF handler
3892 * @family: Address family
3893 * @gifconf: Function handler
3895 * Register protocol dependent address dumping routines. The handler
3896 * that is passed must not be freed or reused until it has been replaced
3897 * by another handler.
3899 int register_gifconf(unsigned int family
, gifconf_func_t
*gifconf
)
3901 if (family
>= NPROTO
)
3903 gifconf_list
[family
] = gifconf
;
3906 EXPORT_SYMBOL(register_gifconf
);
3910 * Map an interface index to its name (SIOCGIFNAME)
3914 * We need this ioctl for efficient implementation of the
3915 * if_indextoname() function required by the IPv6 API. Without
3916 * it, we would have to search all the interfaces to find a
3920 static int dev_ifname(struct net
*net
, struct ifreq __user
*arg
)
3922 struct net_device
*dev
;
3926 * Fetch the caller's info block.
3929 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
3933 dev
= dev_get_by_index_rcu(net
, ifr
.ifr_ifindex
);
3939 strcpy(ifr
.ifr_name
, dev
->name
);
3942 if (copy_to_user(arg
, &ifr
, sizeof(struct ifreq
)))
3948 * Perform a SIOCGIFCONF call. This structure will change
3949 * size eventually, and there is nothing I can do about it.
3950 * Thus we will need a 'compatibility mode'.
3953 static int dev_ifconf(struct net
*net
, char __user
*arg
)
3956 struct net_device
*dev
;
3963 * Fetch the caller's info block.
3966 if (copy_from_user(&ifc
, arg
, sizeof(struct ifconf
)))
3973 * Loop over the interfaces, and write an info block for each.
3977 for_each_netdev(net
, dev
) {
3978 for (i
= 0; i
< NPROTO
; i
++) {
3979 if (gifconf_list
[i
]) {
3982 done
= gifconf_list
[i
](dev
, NULL
, 0);
3984 done
= gifconf_list
[i
](dev
, pos
+ total
,
3994 * All done. Write the updated control block back to the caller.
3996 ifc
.ifc_len
= total
;
3999 * Both BSD and Solaris return 0 here, so we do too.
4001 return copy_to_user(arg
, &ifc
, sizeof(struct ifconf
)) ? -EFAULT
: 0;
4004 #ifdef CONFIG_PROC_FS
4006 #define BUCKET_SPACE (32 - NETDEV_HASHBITS - 1)
4008 #define get_bucket(x) ((x) >> BUCKET_SPACE)
4009 #define get_offset(x) ((x) & ((1 << BUCKET_SPACE) - 1))
4010 #define set_bucket_offset(b, o) ((b) << BUCKET_SPACE | (o))
4012 static inline struct net_device
*dev_from_same_bucket(struct seq_file
*seq
, loff_t
*pos
)
4014 struct net
*net
= seq_file_net(seq
);
4015 struct net_device
*dev
;
4016 struct hlist_node
*p
;
4017 struct hlist_head
*h
;
4018 unsigned int count
= 0, offset
= get_offset(*pos
);
4020 h
= &net
->dev_name_head
[get_bucket(*pos
)];
4021 hlist_for_each_entry_rcu(dev
, p
, h
, name_hlist
) {
4022 if (++count
== offset
)
4029 static inline struct net_device
*dev_from_bucket(struct seq_file
*seq
, loff_t
*pos
)
4031 struct net_device
*dev
;
4032 unsigned int bucket
;
4035 dev
= dev_from_same_bucket(seq
, pos
);
4039 bucket
= get_bucket(*pos
) + 1;
4040 *pos
= set_bucket_offset(bucket
, 1);
4041 } while (bucket
< NETDEV_HASHENTRIES
);
4047 * This is invoked by the /proc filesystem handler to display a device
4050 void *dev_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4055 return SEQ_START_TOKEN
;
4057 if (get_bucket(*pos
) >= NETDEV_HASHENTRIES
)
4060 return dev_from_bucket(seq
, pos
);
4063 void *dev_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4066 return dev_from_bucket(seq
, pos
);
4069 void dev_seq_stop(struct seq_file
*seq
, void *v
)
4075 static void dev_seq_printf_stats(struct seq_file
*seq
, struct net_device
*dev
)
4077 struct rtnl_link_stats64 temp
;
4078 const struct rtnl_link_stats64
*stats
= dev_get_stats(dev
, &temp
);
4080 seq_printf(seq
, "%6s: %7llu %7llu %4llu %4llu %4llu %5llu %10llu %9llu "
4081 "%8llu %7llu %4llu %4llu %4llu %5llu %7llu %10llu\n",
4082 dev
->name
, stats
->rx_bytes
, stats
->rx_packets
,
4084 stats
->rx_dropped
+ stats
->rx_missed_errors
,
4085 stats
->rx_fifo_errors
,
4086 stats
->rx_length_errors
+ stats
->rx_over_errors
+
4087 stats
->rx_crc_errors
+ stats
->rx_frame_errors
,
4088 stats
->rx_compressed
, stats
->multicast
,
4089 stats
->tx_bytes
, stats
->tx_packets
,
4090 stats
->tx_errors
, stats
->tx_dropped
,
4091 stats
->tx_fifo_errors
, stats
->collisions
,
4092 stats
->tx_carrier_errors
+
4093 stats
->tx_aborted_errors
+
4094 stats
->tx_window_errors
+
4095 stats
->tx_heartbeat_errors
,
4096 stats
->tx_compressed
);
4100 * Called from the PROCfs module. This now uses the new arbitrary sized
4101 * /proc/net interface to create /proc/net/dev
4103 static int dev_seq_show(struct seq_file
*seq
, void *v
)
4105 if (v
== SEQ_START_TOKEN
)
4106 seq_puts(seq
, "Inter-| Receive "
4108 " face |bytes packets errs drop fifo frame "
4109 "compressed multicast|bytes packets errs "
4110 "drop fifo colls carrier compressed\n");
4112 dev_seq_printf_stats(seq
, v
);
4116 static struct softnet_data
*softnet_get_online(loff_t
*pos
)
4118 struct softnet_data
*sd
= NULL
;
4120 while (*pos
< nr_cpu_ids
)
4121 if (cpu_online(*pos
)) {
4122 sd
= &per_cpu(softnet_data
, *pos
);
4129 static void *softnet_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4131 return softnet_get_online(pos
);
4134 static void *softnet_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4137 return softnet_get_online(pos
);
4140 static void softnet_seq_stop(struct seq_file
*seq
, void *v
)
4144 static int softnet_seq_show(struct seq_file
*seq
, void *v
)
4146 struct softnet_data
*sd
= v
;
4148 seq_printf(seq
, "%08x %08x %08x %08x %08x %08x %08x %08x %08x %08x\n",
4149 sd
->processed
, sd
->dropped
, sd
->time_squeeze
, 0,
4150 0, 0, 0, 0, /* was fastroute */
4151 sd
->cpu_collision
, sd
->received_rps
);
4155 static const struct seq_operations dev_seq_ops
= {
4156 .start
= dev_seq_start
,
4157 .next
= dev_seq_next
,
4158 .stop
= dev_seq_stop
,
4159 .show
= dev_seq_show
,
4162 static int dev_seq_open(struct inode
*inode
, struct file
*file
)
4164 return seq_open_net(inode
, file
, &dev_seq_ops
,
4165 sizeof(struct seq_net_private
));
4168 static const struct file_operations dev_seq_fops
= {
4169 .owner
= THIS_MODULE
,
4170 .open
= dev_seq_open
,
4172 .llseek
= seq_lseek
,
4173 .release
= seq_release_net
,
4176 static const struct seq_operations softnet_seq_ops
= {
4177 .start
= softnet_seq_start
,
4178 .next
= softnet_seq_next
,
4179 .stop
= softnet_seq_stop
,
4180 .show
= softnet_seq_show
,
4183 static int softnet_seq_open(struct inode
*inode
, struct file
*file
)
4185 return seq_open(file
, &softnet_seq_ops
);
4188 static const struct file_operations softnet_seq_fops
= {
4189 .owner
= THIS_MODULE
,
4190 .open
= softnet_seq_open
,
4192 .llseek
= seq_lseek
,
4193 .release
= seq_release
,
4196 static void *ptype_get_idx(loff_t pos
)
4198 struct packet_type
*pt
= NULL
;
4202 list_for_each_entry_rcu(pt
, &ptype_all
, list
) {
4208 for (t
= 0; t
< PTYPE_HASH_SIZE
; t
++) {
4209 list_for_each_entry_rcu(pt
, &ptype_base
[t
], list
) {
4218 static void *ptype_seq_start(struct seq_file
*seq
, loff_t
*pos
)
4222 return *pos
? ptype_get_idx(*pos
- 1) : SEQ_START_TOKEN
;
4225 static void *ptype_seq_next(struct seq_file
*seq
, void *v
, loff_t
*pos
)
4227 struct packet_type
*pt
;
4228 struct list_head
*nxt
;
4232 if (v
== SEQ_START_TOKEN
)
4233 return ptype_get_idx(0);
4236 nxt
= pt
->list
.next
;
4237 if (pt
->type
== htons(ETH_P_ALL
)) {
4238 if (nxt
!= &ptype_all
)
4241 nxt
= ptype_base
[0].next
;
4243 hash
= ntohs(pt
->type
) & PTYPE_HASH_MASK
;
4245 while (nxt
== &ptype_base
[hash
]) {
4246 if (++hash
>= PTYPE_HASH_SIZE
)
4248 nxt
= ptype_base
[hash
].next
;
4251 return list_entry(nxt
, struct packet_type
, list
);
4254 static void ptype_seq_stop(struct seq_file
*seq
, void *v
)
4260 static int ptype_seq_show(struct seq_file
*seq
, void *v
)
4262 struct packet_type
*pt
= v
;
4264 if (v
== SEQ_START_TOKEN
)
4265 seq_puts(seq
, "Type Device Function\n");
4266 else if (pt
->dev
== NULL
|| dev_net(pt
->dev
) == seq_file_net(seq
)) {
4267 if (pt
->type
== htons(ETH_P_ALL
))
4268 seq_puts(seq
, "ALL ");
4270 seq_printf(seq
, "%04x", ntohs(pt
->type
));
4272 seq_printf(seq
, " %-8s %pF\n",
4273 pt
->dev
? pt
->dev
->name
: "", pt
->func
);
4279 static const struct seq_operations ptype_seq_ops
= {
4280 .start
= ptype_seq_start
,
4281 .next
= ptype_seq_next
,
4282 .stop
= ptype_seq_stop
,
4283 .show
= ptype_seq_show
,
4286 static int ptype_seq_open(struct inode
*inode
, struct file
*file
)
4288 return seq_open_net(inode
, file
, &ptype_seq_ops
,
4289 sizeof(struct seq_net_private
));
4292 static const struct file_operations ptype_seq_fops
= {
4293 .owner
= THIS_MODULE
,
4294 .open
= ptype_seq_open
,
4296 .llseek
= seq_lseek
,
4297 .release
= seq_release_net
,
4301 static int __net_init
dev_proc_net_init(struct net
*net
)
4305 if (!proc_net_fops_create(net
, "dev", S_IRUGO
, &dev_seq_fops
))
4307 if (!proc_net_fops_create(net
, "softnet_stat", S_IRUGO
, &softnet_seq_fops
))
4309 if (!proc_net_fops_create(net
, "ptype", S_IRUGO
, &ptype_seq_fops
))
4312 if (wext_proc_init(net
))
4318 proc_net_remove(net
, "ptype");
4320 proc_net_remove(net
, "softnet_stat");
4322 proc_net_remove(net
, "dev");
4326 static void __net_exit
dev_proc_net_exit(struct net
*net
)
4328 wext_proc_exit(net
);
4330 proc_net_remove(net
, "ptype");
4331 proc_net_remove(net
, "softnet_stat");
4332 proc_net_remove(net
, "dev");
4335 static struct pernet_operations __net_initdata dev_proc_ops
= {
4336 .init
= dev_proc_net_init
,
4337 .exit
= dev_proc_net_exit
,
4340 static int __init
dev_proc_init(void)
4342 return register_pernet_subsys(&dev_proc_ops
);
4345 #define dev_proc_init() 0
4346 #endif /* CONFIG_PROC_FS */
4350 * netdev_set_master - set up master pointer
4351 * @slave: slave device
4352 * @master: new master device
4354 * Changes the master device of the slave. Pass %NULL to break the
4355 * bonding. The caller must hold the RTNL semaphore. On a failure
4356 * a negative errno code is returned. On success the reference counts
4357 * are adjusted and the function returns zero.
4359 int netdev_set_master(struct net_device
*slave
, struct net_device
*master
)
4361 struct net_device
*old
= slave
->master
;
4371 slave
->master
= master
;
4377 EXPORT_SYMBOL(netdev_set_master
);
4380 * netdev_set_bond_master - set up bonding master/slave pair
4381 * @slave: slave device
4382 * @master: new master device
4384 * Changes the master device of the slave. Pass %NULL to break the
4385 * bonding. The caller must hold the RTNL semaphore. On a failure
4386 * a negative errno code is returned. On success %RTM_NEWLINK is sent
4387 * to the routing socket and the function returns zero.
4389 int netdev_set_bond_master(struct net_device
*slave
, struct net_device
*master
)
4395 err
= netdev_set_master(slave
, master
);
4399 slave
->flags
|= IFF_SLAVE
;
4401 slave
->flags
&= ~IFF_SLAVE
;
4403 rtmsg_ifinfo(RTM_NEWLINK
, slave
, IFF_SLAVE
);
4406 EXPORT_SYMBOL(netdev_set_bond_master
);
4408 static void dev_change_rx_flags(struct net_device
*dev
, int flags
)
4410 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4412 if ((dev
->flags
& IFF_UP
) && ops
->ndo_change_rx_flags
)
4413 ops
->ndo_change_rx_flags(dev
, flags
);
4416 static int __dev_set_promiscuity(struct net_device
*dev
, int inc
)
4418 unsigned int old_flags
= dev
->flags
;
4424 dev
->flags
|= IFF_PROMISC
;
4425 dev
->promiscuity
+= inc
;
4426 if (dev
->promiscuity
== 0) {
4429 * If inc causes overflow, untouch promisc and return error.
4432 dev
->flags
&= ~IFF_PROMISC
;
4434 dev
->promiscuity
-= inc
;
4435 pr_warn("%s: promiscuity touches roof, set promiscuity failed. promiscuity feature of device might be broken.\n",
4440 if (dev
->flags
!= old_flags
) {
4441 pr_info("device %s %s promiscuous mode\n",
4443 dev
->flags
& IFF_PROMISC
? "entered" : "left");
4444 if (audit_enabled
) {
4445 current_uid_gid(&uid
, &gid
);
4446 audit_log(current
->audit_context
, GFP_ATOMIC
,
4447 AUDIT_ANOM_PROMISCUOUS
,
4448 "dev=%s prom=%d old_prom=%d auid=%u uid=%u gid=%u ses=%u",
4449 dev
->name
, (dev
->flags
& IFF_PROMISC
),
4450 (old_flags
& IFF_PROMISC
),
4451 audit_get_loginuid(current
),
4453 audit_get_sessionid(current
));
4456 dev_change_rx_flags(dev
, IFF_PROMISC
);
4462 * dev_set_promiscuity - update promiscuity count on a device
4466 * Add or remove promiscuity from a device. While the count in the device
4467 * remains above zero the interface remains promiscuous. Once it hits zero
4468 * the device reverts back to normal filtering operation. A negative inc
4469 * value is used to drop promiscuity on the device.
4470 * Return 0 if successful or a negative errno code on error.
4472 int dev_set_promiscuity(struct net_device
*dev
, int inc
)
4474 unsigned int old_flags
= dev
->flags
;
4477 err
= __dev_set_promiscuity(dev
, inc
);
4480 if (dev
->flags
!= old_flags
)
4481 dev_set_rx_mode(dev
);
4484 EXPORT_SYMBOL(dev_set_promiscuity
);
4487 * dev_set_allmulti - update allmulti count on a device
4491 * Add or remove reception of all multicast frames to a device. While the
4492 * count in the device remains above zero the interface remains listening
4493 * to all interfaces. Once it hits zero the device reverts back to normal
4494 * filtering operation. A negative @inc value is used to drop the counter
4495 * when releasing a resource needing all multicasts.
4496 * Return 0 if successful or a negative errno code on error.
4499 int dev_set_allmulti(struct net_device
*dev
, int inc
)
4501 unsigned int old_flags
= dev
->flags
;
4505 dev
->flags
|= IFF_ALLMULTI
;
4506 dev
->allmulti
+= inc
;
4507 if (dev
->allmulti
== 0) {
4510 * If inc causes overflow, untouch allmulti and return error.
4513 dev
->flags
&= ~IFF_ALLMULTI
;
4515 dev
->allmulti
-= inc
;
4516 pr_warn("%s: allmulti touches roof, set allmulti failed. allmulti feature of device might be broken.\n",
4521 if (dev
->flags
^ old_flags
) {
4522 dev_change_rx_flags(dev
, IFF_ALLMULTI
);
4523 dev_set_rx_mode(dev
);
4527 EXPORT_SYMBOL(dev_set_allmulti
);
4530 * Upload unicast and multicast address lists to device and
4531 * configure RX filtering. When the device doesn't support unicast
4532 * filtering it is put in promiscuous mode while unicast addresses
4535 void __dev_set_rx_mode(struct net_device
*dev
)
4537 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4539 /* dev_open will call this function so the list will stay sane. */
4540 if (!(dev
->flags
&IFF_UP
))
4543 if (!netif_device_present(dev
))
4546 if (!(dev
->priv_flags
& IFF_UNICAST_FLT
)) {
4547 /* Unicast addresses changes may only happen under the rtnl,
4548 * therefore calling __dev_set_promiscuity here is safe.
4550 if (!netdev_uc_empty(dev
) && !dev
->uc_promisc
) {
4551 __dev_set_promiscuity(dev
, 1);
4552 dev
->uc_promisc
= true;
4553 } else if (netdev_uc_empty(dev
) && dev
->uc_promisc
) {
4554 __dev_set_promiscuity(dev
, -1);
4555 dev
->uc_promisc
= false;
4559 if (ops
->ndo_set_rx_mode
)
4560 ops
->ndo_set_rx_mode(dev
);
4563 void dev_set_rx_mode(struct net_device
*dev
)
4565 netif_addr_lock_bh(dev
);
4566 __dev_set_rx_mode(dev
);
4567 netif_addr_unlock_bh(dev
);
4571 * dev_get_flags - get flags reported to userspace
4574 * Get the combination of flag bits exported through APIs to userspace.
4576 unsigned int dev_get_flags(const struct net_device
*dev
)
4580 flags
= (dev
->flags
& ~(IFF_PROMISC
|
4585 (dev
->gflags
& (IFF_PROMISC
|
4588 if (netif_running(dev
)) {
4589 if (netif_oper_up(dev
))
4590 flags
|= IFF_RUNNING
;
4591 if (netif_carrier_ok(dev
))
4592 flags
|= IFF_LOWER_UP
;
4593 if (netif_dormant(dev
))
4594 flags
|= IFF_DORMANT
;
4599 EXPORT_SYMBOL(dev_get_flags
);
4601 int __dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4603 unsigned int old_flags
= dev
->flags
;
4609 * Set the flags on our device.
4612 dev
->flags
= (flags
& (IFF_DEBUG
| IFF_NOTRAILERS
| IFF_NOARP
|
4613 IFF_DYNAMIC
| IFF_MULTICAST
| IFF_PORTSEL
|
4615 (dev
->flags
& (IFF_UP
| IFF_VOLATILE
| IFF_PROMISC
|
4619 * Load in the correct multicast list now the flags have changed.
4622 if ((old_flags
^ flags
) & IFF_MULTICAST
)
4623 dev_change_rx_flags(dev
, IFF_MULTICAST
);
4625 dev_set_rx_mode(dev
);
4628 * Have we downed the interface. We handle IFF_UP ourselves
4629 * according to user attempts to set it, rather than blindly
4634 if ((old_flags
^ flags
) & IFF_UP
) { /* Bit is different ? */
4635 ret
= ((old_flags
& IFF_UP
) ? __dev_close
: __dev_open
)(dev
);
4638 dev_set_rx_mode(dev
);
4641 if ((flags
^ dev
->gflags
) & IFF_PROMISC
) {
4642 int inc
= (flags
& IFF_PROMISC
) ? 1 : -1;
4644 dev
->gflags
^= IFF_PROMISC
;
4645 dev_set_promiscuity(dev
, inc
);
4648 /* NOTE: order of synchronization of IFF_PROMISC and IFF_ALLMULTI
4649 is important. Some (broken) drivers set IFF_PROMISC, when
4650 IFF_ALLMULTI is requested not asking us and not reporting.
4652 if ((flags
^ dev
->gflags
) & IFF_ALLMULTI
) {
4653 int inc
= (flags
& IFF_ALLMULTI
) ? 1 : -1;
4655 dev
->gflags
^= IFF_ALLMULTI
;
4656 dev_set_allmulti(dev
, inc
);
4662 void __dev_notify_flags(struct net_device
*dev
, unsigned int old_flags
)
4664 unsigned int changes
= dev
->flags
^ old_flags
;
4666 if (changes
& IFF_UP
) {
4667 if (dev
->flags
& IFF_UP
)
4668 call_netdevice_notifiers(NETDEV_UP
, dev
);
4670 call_netdevice_notifiers(NETDEV_DOWN
, dev
);
4673 if (dev
->flags
& IFF_UP
&&
4674 (changes
& ~(IFF_UP
| IFF_PROMISC
| IFF_ALLMULTI
| IFF_VOLATILE
)))
4675 call_netdevice_notifiers(NETDEV_CHANGE
, dev
);
4679 * dev_change_flags - change device settings
4681 * @flags: device state flags
4683 * Change settings on device based state flags. The flags are
4684 * in the userspace exported format.
4686 int dev_change_flags(struct net_device
*dev
, unsigned int flags
)
4689 unsigned int changes
, old_flags
= dev
->flags
;
4691 ret
= __dev_change_flags(dev
, flags
);
4695 changes
= old_flags
^ dev
->flags
;
4697 rtmsg_ifinfo(RTM_NEWLINK
, dev
, changes
);
4699 __dev_notify_flags(dev
, old_flags
);
4702 EXPORT_SYMBOL(dev_change_flags
);
4705 * dev_set_mtu - Change maximum transfer unit
4707 * @new_mtu: new transfer unit
4709 * Change the maximum transfer size of the network device.
4711 int dev_set_mtu(struct net_device
*dev
, int new_mtu
)
4713 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4716 if (new_mtu
== dev
->mtu
)
4719 /* MTU must be positive. */
4723 if (!netif_device_present(dev
))
4727 if (ops
->ndo_change_mtu
)
4728 err
= ops
->ndo_change_mtu(dev
, new_mtu
);
4732 if (!err
&& dev
->flags
& IFF_UP
)
4733 call_netdevice_notifiers(NETDEV_CHANGEMTU
, dev
);
4736 EXPORT_SYMBOL(dev_set_mtu
);
4739 * dev_set_group - Change group this device belongs to
4741 * @new_group: group this device should belong to
4743 void dev_set_group(struct net_device
*dev
, int new_group
)
4745 dev
->group
= new_group
;
4747 EXPORT_SYMBOL(dev_set_group
);
4750 * dev_set_mac_address - Change Media Access Control Address
4754 * Change the hardware (MAC) address of the device
4756 int dev_set_mac_address(struct net_device
*dev
, struct sockaddr
*sa
)
4758 const struct net_device_ops
*ops
= dev
->netdev_ops
;
4761 if (!ops
->ndo_set_mac_address
)
4763 if (sa
->sa_family
!= dev
->type
)
4765 if (!netif_device_present(dev
))
4767 err
= ops
->ndo_set_mac_address(dev
, sa
);
4769 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4772 EXPORT_SYMBOL(dev_set_mac_address
);
4775 * Perform the SIOCxIFxxx calls, inside rcu_read_lock()
4777 static int dev_ifsioc_locked(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4780 struct net_device
*dev
= dev_get_by_name_rcu(net
, ifr
->ifr_name
);
4786 case SIOCGIFFLAGS
: /* Get interface flags */
4787 ifr
->ifr_flags
= (short) dev_get_flags(dev
);
4790 case SIOCGIFMETRIC
: /* Get the metric on the interface
4791 (currently unused) */
4792 ifr
->ifr_metric
= 0;
4795 case SIOCGIFMTU
: /* Get the MTU of a device */
4796 ifr
->ifr_mtu
= dev
->mtu
;
4801 memset(ifr
->ifr_hwaddr
.sa_data
, 0, sizeof ifr
->ifr_hwaddr
.sa_data
);
4803 memcpy(ifr
->ifr_hwaddr
.sa_data
, dev
->dev_addr
,
4804 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4805 ifr
->ifr_hwaddr
.sa_family
= dev
->type
;
4813 ifr
->ifr_map
.mem_start
= dev
->mem_start
;
4814 ifr
->ifr_map
.mem_end
= dev
->mem_end
;
4815 ifr
->ifr_map
.base_addr
= dev
->base_addr
;
4816 ifr
->ifr_map
.irq
= dev
->irq
;
4817 ifr
->ifr_map
.dma
= dev
->dma
;
4818 ifr
->ifr_map
.port
= dev
->if_port
;
4822 ifr
->ifr_ifindex
= dev
->ifindex
;
4826 ifr
->ifr_qlen
= dev
->tx_queue_len
;
4830 /* dev_ioctl() should ensure this case
4842 * Perform the SIOCxIFxxx calls, inside rtnl_lock()
4844 static int dev_ifsioc(struct net
*net
, struct ifreq
*ifr
, unsigned int cmd
)
4847 struct net_device
*dev
= __dev_get_by_name(net
, ifr
->ifr_name
);
4848 const struct net_device_ops
*ops
;
4853 ops
= dev
->netdev_ops
;
4856 case SIOCSIFFLAGS
: /* Set interface flags */
4857 return dev_change_flags(dev
, ifr
->ifr_flags
);
4859 case SIOCSIFMETRIC
: /* Set the metric on the interface
4860 (currently unused) */
4863 case SIOCSIFMTU
: /* Set the MTU of a device */
4864 return dev_set_mtu(dev
, ifr
->ifr_mtu
);
4867 return dev_set_mac_address(dev
, &ifr
->ifr_hwaddr
);
4869 case SIOCSIFHWBROADCAST
:
4870 if (ifr
->ifr_hwaddr
.sa_family
!= dev
->type
)
4872 memcpy(dev
->broadcast
, ifr
->ifr_hwaddr
.sa_data
,
4873 min(sizeof ifr
->ifr_hwaddr
.sa_data
, (size_t) dev
->addr_len
));
4874 call_netdevice_notifiers(NETDEV_CHANGEADDR
, dev
);
4878 if (ops
->ndo_set_config
) {
4879 if (!netif_device_present(dev
))
4881 return ops
->ndo_set_config(dev
, &ifr
->ifr_map
);
4886 if (!ops
->ndo_set_rx_mode
||
4887 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4889 if (!netif_device_present(dev
))
4891 return dev_mc_add_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4894 if (!ops
->ndo_set_rx_mode
||
4895 ifr
->ifr_hwaddr
.sa_family
!= AF_UNSPEC
)
4897 if (!netif_device_present(dev
))
4899 return dev_mc_del_global(dev
, ifr
->ifr_hwaddr
.sa_data
);
4902 if (ifr
->ifr_qlen
< 0)
4904 dev
->tx_queue_len
= ifr
->ifr_qlen
;
4908 ifr
->ifr_newname
[IFNAMSIZ
-1] = '\0';
4909 return dev_change_name(dev
, ifr
->ifr_newname
);
4912 err
= net_hwtstamp_validate(ifr
);
4918 * Unknown or private ioctl
4921 if ((cmd
>= SIOCDEVPRIVATE
&&
4922 cmd
<= SIOCDEVPRIVATE
+ 15) ||
4923 cmd
== SIOCBONDENSLAVE
||
4924 cmd
== SIOCBONDRELEASE
||
4925 cmd
== SIOCBONDSETHWADDR
||
4926 cmd
== SIOCBONDSLAVEINFOQUERY
||
4927 cmd
== SIOCBONDINFOQUERY
||
4928 cmd
== SIOCBONDCHANGEACTIVE
||
4929 cmd
== SIOCGMIIPHY
||
4930 cmd
== SIOCGMIIREG
||
4931 cmd
== SIOCSMIIREG
||
4932 cmd
== SIOCBRADDIF
||
4933 cmd
== SIOCBRDELIF
||
4934 cmd
== SIOCSHWTSTAMP
||
4935 cmd
== SIOCWANDEV
) {
4937 if (ops
->ndo_do_ioctl
) {
4938 if (netif_device_present(dev
))
4939 err
= ops
->ndo_do_ioctl(dev
, ifr
, cmd
);
4951 * This function handles all "interface"-type I/O control requests. The actual
4952 * 'doing' part of this is dev_ifsioc above.
4956 * dev_ioctl - network device ioctl
4957 * @net: the applicable net namespace
4958 * @cmd: command to issue
4959 * @arg: pointer to a struct ifreq in user space
4961 * Issue ioctl functions to devices. This is normally called by the
4962 * user space syscall interfaces but can sometimes be useful for
4963 * other purposes. The return value is the return from the syscall if
4964 * positive or a negative errno code on error.
4967 int dev_ioctl(struct net
*net
, unsigned int cmd
, void __user
*arg
)
4973 /* One special case: SIOCGIFCONF takes ifconf argument
4974 and requires shared lock, because it sleeps writing
4978 if (cmd
== SIOCGIFCONF
) {
4980 ret
= dev_ifconf(net
, (char __user
*) arg
);
4984 if (cmd
== SIOCGIFNAME
)
4985 return dev_ifname(net
, (struct ifreq __user
*)arg
);
4987 if (copy_from_user(&ifr
, arg
, sizeof(struct ifreq
)))
4990 ifr
.ifr_name
[IFNAMSIZ
-1] = 0;
4992 colon
= strchr(ifr
.ifr_name
, ':');
4997 * See which interface the caller is talking about.
5002 * These ioctl calls:
5003 * - can be done by all.
5004 * - atomic and do not require locking.
5015 dev_load(net
, ifr
.ifr_name
);
5017 ret
= dev_ifsioc_locked(net
, &ifr
, cmd
);
5022 if (copy_to_user(arg
, &ifr
,
5023 sizeof(struct ifreq
)))
5029 dev_load(net
, ifr
.ifr_name
);
5031 ret
= dev_ethtool(net
, &ifr
);
5036 if (copy_to_user(arg
, &ifr
,
5037 sizeof(struct ifreq
)))
5043 * These ioctl calls:
5044 * - require superuser power.
5045 * - require strict serialization.
5051 if (!capable(CAP_NET_ADMIN
))
5053 dev_load(net
, ifr
.ifr_name
);
5055 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5060 if (copy_to_user(arg
, &ifr
,
5061 sizeof(struct ifreq
)))
5067 * These ioctl calls:
5068 * - require superuser power.
5069 * - require strict serialization.
5070 * - do not return a value
5080 case SIOCSIFHWBROADCAST
:
5083 case SIOCBONDENSLAVE
:
5084 case SIOCBONDRELEASE
:
5085 case SIOCBONDSETHWADDR
:
5086 case SIOCBONDCHANGEACTIVE
:
5090 if (!capable(CAP_NET_ADMIN
))
5093 case SIOCBONDSLAVEINFOQUERY
:
5094 case SIOCBONDINFOQUERY
:
5095 dev_load(net
, ifr
.ifr_name
);
5097 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5102 /* Get the per device memory space. We can add this but
5103 * currently do not support it */
5105 /* Set the per device memory buffer space.
5106 * Not applicable in our case */
5111 * Unknown or private ioctl.
5114 if (cmd
== SIOCWANDEV
||
5115 (cmd
>= SIOCDEVPRIVATE
&&
5116 cmd
<= SIOCDEVPRIVATE
+ 15)) {
5117 dev_load(net
, ifr
.ifr_name
);
5119 ret
= dev_ifsioc(net
, &ifr
, cmd
);
5121 if (!ret
&& copy_to_user(arg
, &ifr
,
5122 sizeof(struct ifreq
)))
5126 /* Take care of Wireless Extensions */
5127 if (cmd
>= SIOCIWFIRST
&& cmd
<= SIOCIWLAST
)
5128 return wext_handle_ioctl(net
, &ifr
, cmd
, arg
);
5135 * dev_new_index - allocate an ifindex
5136 * @net: the applicable net namespace
5138 * Returns a suitable unique value for a new device interface
5139 * number. The caller must hold the rtnl semaphore or the
5140 * dev_base_lock to be sure it remains unique.
5142 static int dev_new_index(struct net
*net
)
5148 if (!__dev_get_by_index(net
, ifindex
))
5153 /* Delayed registration/unregisteration */
5154 static LIST_HEAD(net_todo_list
);
5156 static void net_set_todo(struct net_device
*dev
)
5158 list_add_tail(&dev
->todo_list
, &net_todo_list
);
5161 static void rollback_registered_many(struct list_head
*head
)
5163 struct net_device
*dev
, *tmp
;
5165 BUG_ON(dev_boot_phase
);
5168 list_for_each_entry_safe(dev
, tmp
, head
, unreg_list
) {
5169 /* Some devices call without registering
5170 * for initialization unwind. Remove those
5171 * devices and proceed with the remaining.
5173 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5174 pr_debug("unregister_netdevice: device %s/%p never was registered\n",
5178 list_del(&dev
->unreg_list
);
5181 dev
->dismantle
= true;
5182 BUG_ON(dev
->reg_state
!= NETREG_REGISTERED
);
5185 /* If device is running, close it first. */
5186 dev_close_many(head
);
5188 list_for_each_entry(dev
, head
, unreg_list
) {
5189 /* And unlink it from device chain. */
5190 unlist_netdevice(dev
);
5192 dev
->reg_state
= NETREG_UNREGISTERING
;
5197 list_for_each_entry(dev
, head
, unreg_list
) {
5198 /* Shutdown queueing discipline. */
5202 /* Notify protocols, that we are about to destroy
5203 this device. They should clean all the things.
5205 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5207 if (!dev
->rtnl_link_ops
||
5208 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5209 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
5212 * Flush the unicast and multicast chains
5217 if (dev
->netdev_ops
->ndo_uninit
)
5218 dev
->netdev_ops
->ndo_uninit(dev
);
5220 /* Notifier chain MUST detach us from master device. */
5221 WARN_ON(dev
->master
);
5223 /* Remove entries from kobject tree */
5224 netdev_unregister_kobject(dev
);
5227 /* Process any work delayed until the end of the batch */
5228 dev
= list_first_entry(head
, struct net_device
, unreg_list
);
5229 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
5233 list_for_each_entry(dev
, head
, unreg_list
)
5237 static void rollback_registered(struct net_device
*dev
)
5241 list_add(&dev
->unreg_list
, &single
);
5242 rollback_registered_many(&single
);
5246 static netdev_features_t
netdev_fix_features(struct net_device
*dev
,
5247 netdev_features_t features
)
5249 /* Fix illegal checksum combinations */
5250 if ((features
& NETIF_F_HW_CSUM
) &&
5251 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5252 netdev_warn(dev
, "mixed HW and IP checksum settings.\n");
5253 features
&= ~(NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
);
5256 /* Fix illegal SG+CSUM combinations. */
5257 if ((features
& NETIF_F_SG
) &&
5258 !(features
& NETIF_F_ALL_CSUM
)) {
5260 "Dropping NETIF_F_SG since no checksum feature.\n");
5261 features
&= ~NETIF_F_SG
;
5264 /* TSO requires that SG is present as well. */
5265 if ((features
& NETIF_F_ALL_TSO
) && !(features
& NETIF_F_SG
)) {
5266 netdev_dbg(dev
, "Dropping TSO features since no SG feature.\n");
5267 features
&= ~NETIF_F_ALL_TSO
;
5270 /* TSO ECN requires that TSO is present as well. */
5271 if ((features
& NETIF_F_ALL_TSO
) == NETIF_F_TSO_ECN
)
5272 features
&= ~NETIF_F_TSO_ECN
;
5274 /* Software GSO depends on SG. */
5275 if ((features
& NETIF_F_GSO
) && !(features
& NETIF_F_SG
)) {
5276 netdev_dbg(dev
, "Dropping NETIF_F_GSO since no SG feature.\n");
5277 features
&= ~NETIF_F_GSO
;
5280 /* UFO needs SG and checksumming */
5281 if (features
& NETIF_F_UFO
) {
5282 /* maybe split UFO into V4 and V6? */
5283 if (!((features
& NETIF_F_GEN_CSUM
) ||
5284 (features
& (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))
5285 == (NETIF_F_IP_CSUM
|NETIF_F_IPV6_CSUM
))) {
5287 "Dropping NETIF_F_UFO since no checksum offload features.\n");
5288 features
&= ~NETIF_F_UFO
;
5291 if (!(features
& NETIF_F_SG
)) {
5293 "Dropping NETIF_F_UFO since no NETIF_F_SG feature.\n");
5294 features
&= ~NETIF_F_UFO
;
5301 int __netdev_update_features(struct net_device
*dev
)
5303 netdev_features_t features
;
5308 features
= netdev_get_wanted_features(dev
);
5310 if (dev
->netdev_ops
->ndo_fix_features
)
5311 features
= dev
->netdev_ops
->ndo_fix_features(dev
, features
);
5313 /* driver might be less strict about feature dependencies */
5314 features
= netdev_fix_features(dev
, features
);
5316 if (dev
->features
== features
)
5319 netdev_dbg(dev
, "Features changed: %pNF -> %pNF\n",
5320 &dev
->features
, &features
);
5322 if (dev
->netdev_ops
->ndo_set_features
)
5323 err
= dev
->netdev_ops
->ndo_set_features(dev
, features
);
5325 if (unlikely(err
< 0)) {
5327 "set_features() failed (%d); wanted %pNF, left %pNF\n",
5328 err
, &features
, &dev
->features
);
5333 dev
->features
= features
;
5339 * netdev_update_features - recalculate device features
5340 * @dev: the device to check
5342 * Recalculate dev->features set and send notifications if it
5343 * has changed. Should be called after driver or hardware dependent
5344 * conditions might have changed that influence the features.
5346 void netdev_update_features(struct net_device
*dev
)
5348 if (__netdev_update_features(dev
))
5349 netdev_features_change(dev
);
5351 EXPORT_SYMBOL(netdev_update_features
);
5354 * netdev_change_features - recalculate device features
5355 * @dev: the device to check
5357 * Recalculate dev->features set and send notifications even
5358 * if they have not changed. Should be called instead of
5359 * netdev_update_features() if also dev->vlan_features might
5360 * have changed to allow the changes to be propagated to stacked
5363 void netdev_change_features(struct net_device
*dev
)
5365 __netdev_update_features(dev
);
5366 netdev_features_change(dev
);
5368 EXPORT_SYMBOL(netdev_change_features
);
5371 * netif_stacked_transfer_operstate - transfer operstate
5372 * @rootdev: the root or lower level device to transfer state from
5373 * @dev: the device to transfer operstate to
5375 * Transfer operational state from root to device. This is normally
5376 * called when a stacking relationship exists between the root
5377 * device and the device(a leaf device).
5379 void netif_stacked_transfer_operstate(const struct net_device
*rootdev
,
5380 struct net_device
*dev
)
5382 if (rootdev
->operstate
== IF_OPER_DORMANT
)
5383 netif_dormant_on(dev
);
5385 netif_dormant_off(dev
);
5387 if (netif_carrier_ok(rootdev
)) {
5388 if (!netif_carrier_ok(dev
))
5389 netif_carrier_on(dev
);
5391 if (netif_carrier_ok(dev
))
5392 netif_carrier_off(dev
);
5395 EXPORT_SYMBOL(netif_stacked_transfer_operstate
);
5398 static int netif_alloc_rx_queues(struct net_device
*dev
)
5400 unsigned int i
, count
= dev
->num_rx_queues
;
5401 struct netdev_rx_queue
*rx
;
5405 rx
= kcalloc(count
, sizeof(struct netdev_rx_queue
), GFP_KERNEL
);
5407 pr_err("netdev: Unable to allocate %u rx queues\n", count
);
5412 for (i
= 0; i
< count
; i
++)
5418 static void netdev_init_one_queue(struct net_device
*dev
,
5419 struct netdev_queue
*queue
, void *_unused
)
5421 /* Initialize queue lock */
5422 spin_lock_init(&queue
->_xmit_lock
);
5423 netdev_set_xmit_lockdep_class(&queue
->_xmit_lock
, dev
->type
);
5424 queue
->xmit_lock_owner
= -1;
5425 netdev_queue_numa_node_write(queue
, NUMA_NO_NODE
);
5428 dql_init(&queue
->dql
, HZ
);
5432 static int netif_alloc_netdev_queues(struct net_device
*dev
)
5434 unsigned int count
= dev
->num_tx_queues
;
5435 struct netdev_queue
*tx
;
5439 tx
= kcalloc(count
, sizeof(struct netdev_queue
), GFP_KERNEL
);
5441 pr_err("netdev: Unable to allocate %u tx queues\n", count
);
5446 netdev_for_each_tx_queue(dev
, netdev_init_one_queue
, NULL
);
5447 spin_lock_init(&dev
->tx_global_lock
);
5453 * register_netdevice - register a network device
5454 * @dev: device to register
5456 * Take a completed network device structure and add it to the kernel
5457 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5458 * chain. 0 is returned on success. A negative errno code is returned
5459 * on a failure to set up the device, or if the name is a duplicate.
5461 * Callers must hold the rtnl semaphore. You may want
5462 * register_netdev() instead of this.
5465 * The locking appears insufficient to guarantee two parallel registers
5466 * will not get the same name.
5469 int register_netdevice(struct net_device
*dev
)
5472 struct net
*net
= dev_net(dev
);
5474 BUG_ON(dev_boot_phase
);
5479 /* When net_device's are persistent, this will be fatal. */
5480 BUG_ON(dev
->reg_state
!= NETREG_UNINITIALIZED
);
5483 spin_lock_init(&dev
->addr_list_lock
);
5484 netdev_set_addr_lockdep_class(dev
);
5488 ret
= dev_get_valid_name(dev
, dev
->name
);
5492 /* Init, if this function is available */
5493 if (dev
->netdev_ops
->ndo_init
) {
5494 ret
= dev
->netdev_ops
->ndo_init(dev
);
5502 dev
->ifindex
= dev_new_index(net
);
5503 if (dev
->iflink
== -1)
5504 dev
->iflink
= dev
->ifindex
;
5506 /* Transfer changeable features to wanted_features and enable
5507 * software offloads (GSO and GRO).
5509 dev
->hw_features
|= NETIF_F_SOFT_FEATURES
;
5510 dev
->features
|= NETIF_F_SOFT_FEATURES
;
5511 dev
->wanted_features
= dev
->features
& dev
->hw_features
;
5513 /* Turn on no cache copy if HW is doing checksum */
5514 if (!(dev
->flags
& IFF_LOOPBACK
)) {
5515 dev
->hw_features
|= NETIF_F_NOCACHE_COPY
;
5516 if (dev
->features
& NETIF_F_ALL_CSUM
) {
5517 dev
->wanted_features
|= NETIF_F_NOCACHE_COPY
;
5518 dev
->features
|= NETIF_F_NOCACHE_COPY
;
5522 /* Make NETIF_F_HIGHDMA inheritable to VLAN devices.
5524 dev
->vlan_features
|= NETIF_F_HIGHDMA
;
5526 ret
= call_netdevice_notifiers(NETDEV_POST_INIT
, dev
);
5527 ret
= notifier_to_errno(ret
);
5531 ret
= netdev_register_kobject(dev
);
5534 dev
->reg_state
= NETREG_REGISTERED
;
5536 __netdev_update_features(dev
);
5539 * Default initial state at registry is that the
5540 * device is present.
5543 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5545 dev_init_scheduler(dev
);
5547 list_netdevice(dev
);
5549 /* Notify protocols, that a new device appeared. */
5550 ret
= call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
5551 ret
= notifier_to_errno(ret
);
5553 rollback_registered(dev
);
5554 dev
->reg_state
= NETREG_UNREGISTERED
;
5557 * Prevent userspace races by waiting until the network
5558 * device is fully setup before sending notifications.
5560 if (!dev
->rtnl_link_ops
||
5561 dev
->rtnl_link_state
== RTNL_LINK_INITIALIZED
)
5562 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
5568 if (dev
->netdev_ops
->ndo_uninit
)
5569 dev
->netdev_ops
->ndo_uninit(dev
);
5572 EXPORT_SYMBOL(register_netdevice
);
5575 * init_dummy_netdev - init a dummy network device for NAPI
5576 * @dev: device to init
5578 * This takes a network device structure and initialize the minimum
5579 * amount of fields so it can be used to schedule NAPI polls without
5580 * registering a full blown interface. This is to be used by drivers
5581 * that need to tie several hardware interfaces to a single NAPI
5582 * poll scheduler due to HW limitations.
5584 int init_dummy_netdev(struct net_device
*dev
)
5586 /* Clear everything. Note we don't initialize spinlocks
5587 * are they aren't supposed to be taken by any of the
5588 * NAPI code and this dummy netdev is supposed to be
5589 * only ever used for NAPI polls
5591 memset(dev
, 0, sizeof(struct net_device
));
5593 /* make sure we BUG if trying to hit standard
5594 * register/unregister code path
5596 dev
->reg_state
= NETREG_DUMMY
;
5598 /* NAPI wants this */
5599 INIT_LIST_HEAD(&dev
->napi_list
);
5601 /* a dummy interface is started by default */
5602 set_bit(__LINK_STATE_PRESENT
, &dev
->state
);
5603 set_bit(__LINK_STATE_START
, &dev
->state
);
5605 /* Note : We dont allocate pcpu_refcnt for dummy devices,
5606 * because users of this 'device' dont need to change
5612 EXPORT_SYMBOL_GPL(init_dummy_netdev
);
5616 * register_netdev - register a network device
5617 * @dev: device to register
5619 * Take a completed network device structure and add it to the kernel
5620 * interfaces. A %NETDEV_REGISTER message is sent to the netdev notifier
5621 * chain. 0 is returned on success. A negative errno code is returned
5622 * on a failure to set up the device, or if the name is a duplicate.
5624 * This is a wrapper around register_netdevice that takes the rtnl semaphore
5625 * and expands the device name if you passed a format string to
5628 int register_netdev(struct net_device
*dev
)
5633 err
= register_netdevice(dev
);
5637 EXPORT_SYMBOL(register_netdev
);
5639 int netdev_refcnt_read(const struct net_device
*dev
)
5643 for_each_possible_cpu(i
)
5644 refcnt
+= *per_cpu_ptr(dev
->pcpu_refcnt
, i
);
5647 EXPORT_SYMBOL(netdev_refcnt_read
);
5650 * netdev_wait_allrefs - wait until all references are gone.
5652 * This is called when unregistering network devices.
5654 * Any protocol or device that holds a reference should register
5655 * for netdevice notification, and cleanup and put back the
5656 * reference if they receive an UNREGISTER event.
5657 * We can get stuck here if buggy protocols don't correctly
5660 static void netdev_wait_allrefs(struct net_device
*dev
)
5662 unsigned long rebroadcast_time
, warning_time
;
5665 linkwatch_forget_dev(dev
);
5667 rebroadcast_time
= warning_time
= jiffies
;
5668 refcnt
= netdev_refcnt_read(dev
);
5670 while (refcnt
!= 0) {
5671 if (time_after(jiffies
, rebroadcast_time
+ 1 * HZ
)) {
5674 /* Rebroadcast unregister notification */
5675 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
5676 /* don't resend NETDEV_UNREGISTER_BATCH, _BATCH users
5677 * should have already handle it the first time */
5679 if (test_bit(__LINK_STATE_LINKWATCH_PENDING
,
5681 /* We must not have linkwatch events
5682 * pending on unregister. If this
5683 * happens, we simply run the queue
5684 * unscheduled, resulting in a noop
5687 linkwatch_run_queue();
5692 rebroadcast_time
= jiffies
;
5697 refcnt
= netdev_refcnt_read(dev
);
5699 if (time_after(jiffies
, warning_time
+ 10 * HZ
)) {
5700 pr_emerg("unregister_netdevice: waiting for %s to become free. Usage count = %d\n",
5702 warning_time
= jiffies
;
5711 * register_netdevice(x1);
5712 * register_netdevice(x2);
5714 * unregister_netdevice(y1);
5715 * unregister_netdevice(y2);
5721 * We are invoked by rtnl_unlock().
5722 * This allows us to deal with problems:
5723 * 1) We can delete sysfs objects which invoke hotplug
5724 * without deadlocking with linkwatch via keventd.
5725 * 2) Since we run with the RTNL semaphore not held, we can sleep
5726 * safely in order to wait for the netdev refcnt to drop to zero.
5728 * We must not return until all unregister events added during
5729 * the interval the lock was held have been completed.
5731 void netdev_run_todo(void)
5733 struct list_head list
;
5735 /* Snapshot list, allow later requests */
5736 list_replace_init(&net_todo_list
, &list
);
5740 /* Wait for rcu callbacks to finish before attempting to drain
5741 * the device list. This usually avoids a 250ms wait.
5743 if (!list_empty(&list
))
5746 while (!list_empty(&list
)) {
5747 struct net_device
*dev
5748 = list_first_entry(&list
, struct net_device
, todo_list
);
5749 list_del(&dev
->todo_list
);
5751 if (unlikely(dev
->reg_state
!= NETREG_UNREGISTERING
)) {
5752 pr_err("network todo '%s' but state %d\n",
5753 dev
->name
, dev
->reg_state
);
5758 dev
->reg_state
= NETREG_UNREGISTERED
;
5760 on_each_cpu(flush_backlog
, dev
, 1);
5762 netdev_wait_allrefs(dev
);
5765 BUG_ON(netdev_refcnt_read(dev
));
5766 WARN_ON(rcu_access_pointer(dev
->ip_ptr
));
5767 WARN_ON(rcu_access_pointer(dev
->ip6_ptr
));
5768 WARN_ON(dev
->dn_ptr
);
5770 if (dev
->destructor
)
5771 dev
->destructor(dev
);
5773 /* Free network device */
5774 kobject_put(&dev
->dev
.kobj
);
5778 /* Convert net_device_stats to rtnl_link_stats64. They have the same
5779 * fields in the same order, with only the type differing.
5781 void netdev_stats_to_stats64(struct rtnl_link_stats64
*stats64
,
5782 const struct net_device_stats
*netdev_stats
)
5784 #if BITS_PER_LONG == 64
5785 BUILD_BUG_ON(sizeof(*stats64
) != sizeof(*netdev_stats
));
5786 memcpy(stats64
, netdev_stats
, sizeof(*stats64
));
5788 size_t i
, n
= sizeof(*stats64
) / sizeof(u64
);
5789 const unsigned long *src
= (const unsigned long *)netdev_stats
;
5790 u64
*dst
= (u64
*)stats64
;
5792 BUILD_BUG_ON(sizeof(*netdev_stats
) / sizeof(unsigned long) !=
5793 sizeof(*stats64
) / sizeof(u64
));
5794 for (i
= 0; i
< n
; i
++)
5798 EXPORT_SYMBOL(netdev_stats_to_stats64
);
5801 * dev_get_stats - get network device statistics
5802 * @dev: device to get statistics from
5803 * @storage: place to store stats
5805 * Get network statistics from device. Return @storage.
5806 * The device driver may provide its own method by setting
5807 * dev->netdev_ops->get_stats64 or dev->netdev_ops->get_stats;
5808 * otherwise the internal statistics structure is used.
5810 struct rtnl_link_stats64
*dev_get_stats(struct net_device
*dev
,
5811 struct rtnl_link_stats64
*storage
)
5813 const struct net_device_ops
*ops
= dev
->netdev_ops
;
5815 if (ops
->ndo_get_stats64
) {
5816 memset(storage
, 0, sizeof(*storage
));
5817 ops
->ndo_get_stats64(dev
, storage
);
5818 } else if (ops
->ndo_get_stats
) {
5819 netdev_stats_to_stats64(storage
, ops
->ndo_get_stats(dev
));
5821 netdev_stats_to_stats64(storage
, &dev
->stats
);
5823 storage
->rx_dropped
+= atomic_long_read(&dev
->rx_dropped
);
5826 EXPORT_SYMBOL(dev_get_stats
);
5828 struct netdev_queue
*dev_ingress_queue_create(struct net_device
*dev
)
5830 struct netdev_queue
*queue
= dev_ingress_queue(dev
);
5832 #ifdef CONFIG_NET_CLS_ACT
5835 queue
= kzalloc(sizeof(*queue
), GFP_KERNEL
);
5838 netdev_init_one_queue(dev
, queue
, NULL
);
5839 queue
->qdisc
= &noop_qdisc
;
5840 queue
->qdisc_sleeping
= &noop_qdisc
;
5841 rcu_assign_pointer(dev
->ingress_queue
, queue
);
5847 * alloc_netdev_mqs - allocate network device
5848 * @sizeof_priv: size of private data to allocate space for
5849 * @name: device name format string
5850 * @setup: callback to initialize device
5851 * @txqs: the number of TX subqueues to allocate
5852 * @rxqs: the number of RX subqueues to allocate
5854 * Allocates a struct net_device with private data area for driver use
5855 * and performs basic initialization. Also allocates subquue structs
5856 * for each queue on the device.
5858 struct net_device
*alloc_netdev_mqs(int sizeof_priv
, const char *name
,
5859 void (*setup
)(struct net_device
*),
5860 unsigned int txqs
, unsigned int rxqs
)
5862 struct net_device
*dev
;
5864 struct net_device
*p
;
5866 BUG_ON(strlen(name
) >= sizeof(dev
->name
));
5869 pr_err("alloc_netdev: Unable to allocate device with zero queues\n");
5875 pr_err("alloc_netdev: Unable to allocate device with zero RX queues\n");
5880 alloc_size
= sizeof(struct net_device
);
5882 /* ensure 32-byte alignment of private area */
5883 alloc_size
= ALIGN(alloc_size
, NETDEV_ALIGN
);
5884 alloc_size
+= sizeof_priv
;
5886 /* ensure 32-byte alignment of whole construct */
5887 alloc_size
+= NETDEV_ALIGN
- 1;
5889 p
= kzalloc(alloc_size
, GFP_KERNEL
);
5891 pr_err("alloc_netdev: Unable to allocate device\n");
5895 dev
= PTR_ALIGN(p
, NETDEV_ALIGN
);
5896 dev
->padded
= (char *)dev
- (char *)p
;
5898 dev
->pcpu_refcnt
= alloc_percpu(int);
5899 if (!dev
->pcpu_refcnt
)
5902 if (dev_addr_init(dev
))
5908 dev_net_set(dev
, &init_net
);
5910 dev
->gso_max_size
= GSO_MAX_SIZE
;
5912 INIT_LIST_HEAD(&dev
->napi_list
);
5913 INIT_LIST_HEAD(&dev
->unreg_list
);
5914 INIT_LIST_HEAD(&dev
->link_watch_list
);
5915 dev
->priv_flags
= IFF_XMIT_DST_RELEASE
;
5918 dev
->num_tx_queues
= txqs
;
5919 dev
->real_num_tx_queues
= txqs
;
5920 if (netif_alloc_netdev_queues(dev
))
5924 dev
->num_rx_queues
= rxqs
;
5925 dev
->real_num_rx_queues
= rxqs
;
5926 if (netif_alloc_rx_queues(dev
))
5930 strcpy(dev
->name
, name
);
5931 dev
->group
= INIT_NETDEV_GROUP
;
5939 free_percpu(dev
->pcpu_refcnt
);
5949 EXPORT_SYMBOL(alloc_netdev_mqs
);
5952 * free_netdev - free network device
5955 * This function does the last stage of destroying an allocated device
5956 * interface. The reference to the device object is released.
5957 * If this is the last reference then it will be freed.
5959 void free_netdev(struct net_device
*dev
)
5961 struct napi_struct
*p
, *n
;
5963 release_net(dev_net(dev
));
5970 kfree(rcu_dereference_protected(dev
->ingress_queue
, 1));
5972 /* Flush device addresses */
5973 dev_addr_flush(dev
);
5975 list_for_each_entry_safe(p
, n
, &dev
->napi_list
, dev_list
)
5978 free_percpu(dev
->pcpu_refcnt
);
5979 dev
->pcpu_refcnt
= NULL
;
5981 /* Compatibility with error handling in drivers */
5982 if (dev
->reg_state
== NETREG_UNINITIALIZED
) {
5983 kfree((char *)dev
- dev
->padded
);
5987 BUG_ON(dev
->reg_state
!= NETREG_UNREGISTERED
);
5988 dev
->reg_state
= NETREG_RELEASED
;
5990 /* will free via device release */
5991 put_device(&dev
->dev
);
5993 EXPORT_SYMBOL(free_netdev
);
5996 * synchronize_net - Synchronize with packet receive processing
5998 * Wait for packets currently being received to be done.
5999 * Does not block later packets from starting.
6001 void synchronize_net(void)
6004 if (rtnl_is_locked())
6005 synchronize_rcu_expedited();
6009 EXPORT_SYMBOL(synchronize_net
);
6012 * unregister_netdevice_queue - remove device from the kernel
6016 * This function shuts down a device interface and removes it
6017 * from the kernel tables.
6018 * If head not NULL, device is queued to be unregistered later.
6020 * Callers must hold the rtnl semaphore. You may want
6021 * unregister_netdev() instead of this.
6024 void unregister_netdevice_queue(struct net_device
*dev
, struct list_head
*head
)
6029 list_move_tail(&dev
->unreg_list
, head
);
6031 rollback_registered(dev
);
6032 /* Finish processing unregister after unlock */
6036 EXPORT_SYMBOL(unregister_netdevice_queue
);
6039 * unregister_netdevice_many - unregister many devices
6040 * @head: list of devices
6042 void unregister_netdevice_many(struct list_head
*head
)
6044 struct net_device
*dev
;
6046 if (!list_empty(head
)) {
6047 rollback_registered_many(head
);
6048 list_for_each_entry(dev
, head
, unreg_list
)
6052 EXPORT_SYMBOL(unregister_netdevice_many
);
6055 * unregister_netdev - remove device from the kernel
6058 * This function shuts down a device interface and removes it
6059 * from the kernel tables.
6061 * This is just a wrapper for unregister_netdevice that takes
6062 * the rtnl semaphore. In general you want to use this and not
6063 * unregister_netdevice.
6065 void unregister_netdev(struct net_device
*dev
)
6068 unregister_netdevice(dev
);
6071 EXPORT_SYMBOL(unregister_netdev
);
6074 * dev_change_net_namespace - move device to different nethost namespace
6076 * @net: network namespace
6077 * @pat: If not NULL name pattern to try if the current device name
6078 * is already taken in the destination network namespace.
6080 * This function shuts down a device interface and moves it
6081 * to a new network namespace. On success 0 is returned, on
6082 * a failure a netagive errno code is returned.
6084 * Callers must hold the rtnl semaphore.
6087 int dev_change_net_namespace(struct net_device
*dev
, struct net
*net
, const char *pat
)
6093 /* Don't allow namespace local devices to be moved. */
6095 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6098 /* Ensure the device has been registrered */
6100 if (dev
->reg_state
!= NETREG_REGISTERED
)
6103 /* Get out if there is nothing todo */
6105 if (net_eq(dev_net(dev
), net
))
6108 /* Pick the destination device name, and ensure
6109 * we can use it in the destination network namespace.
6112 if (__dev_get_by_name(net
, dev
->name
)) {
6113 /* We get here if we can't use the current device name */
6116 if (dev_get_valid_name(dev
, pat
) < 0)
6121 * And now a mini version of register_netdevice unregister_netdevice.
6124 /* If device is running close it first. */
6127 /* And unlink it from device chain */
6129 unlist_netdevice(dev
);
6133 /* Shutdown queueing discipline. */
6136 /* Notify protocols, that we are about to destroy
6137 this device. They should clean all the things.
6139 Note that dev->reg_state stays at NETREG_REGISTERED.
6140 This is wanted because this way 8021q and macvlan know
6141 the device is just moving and can keep their slaves up.
6143 call_netdevice_notifiers(NETDEV_UNREGISTER
, dev
);
6144 call_netdevice_notifiers(NETDEV_UNREGISTER_BATCH
, dev
);
6145 rtmsg_ifinfo(RTM_DELLINK
, dev
, ~0U);
6148 * Flush the unicast and multicast chains
6153 /* Actually switch the network namespace */
6154 dev_net_set(dev
, net
);
6156 /* If there is an ifindex conflict assign a new one */
6157 if (__dev_get_by_index(net
, dev
->ifindex
)) {
6158 int iflink
= (dev
->iflink
== dev
->ifindex
);
6159 dev
->ifindex
= dev_new_index(net
);
6161 dev
->iflink
= dev
->ifindex
;
6164 /* Fixup kobjects */
6165 err
= device_rename(&dev
->dev
, dev
->name
);
6168 /* Add the device back in the hashes */
6169 list_netdevice(dev
);
6171 /* Notify protocols, that a new device appeared. */
6172 call_netdevice_notifiers(NETDEV_REGISTER
, dev
);
6175 * Prevent userspace races by waiting until the network
6176 * device is fully setup before sending notifications.
6178 rtmsg_ifinfo(RTM_NEWLINK
, dev
, ~0U);
6185 EXPORT_SYMBOL_GPL(dev_change_net_namespace
);
6187 static int dev_cpu_callback(struct notifier_block
*nfb
,
6188 unsigned long action
,
6191 struct sk_buff
**list_skb
;
6192 struct sk_buff
*skb
;
6193 unsigned int cpu
, oldcpu
= (unsigned long)ocpu
;
6194 struct softnet_data
*sd
, *oldsd
;
6196 if (action
!= CPU_DEAD
&& action
!= CPU_DEAD_FROZEN
)
6199 local_irq_disable();
6200 cpu
= smp_processor_id();
6201 sd
= &per_cpu(softnet_data
, cpu
);
6202 oldsd
= &per_cpu(softnet_data
, oldcpu
);
6204 /* Find end of our completion_queue. */
6205 list_skb
= &sd
->completion_queue
;
6207 list_skb
= &(*list_skb
)->next
;
6208 /* Append completion queue from offline CPU. */
6209 *list_skb
= oldsd
->completion_queue
;
6210 oldsd
->completion_queue
= NULL
;
6212 /* Append output queue from offline CPU. */
6213 if (oldsd
->output_queue
) {
6214 *sd
->output_queue_tailp
= oldsd
->output_queue
;
6215 sd
->output_queue_tailp
= oldsd
->output_queue_tailp
;
6216 oldsd
->output_queue
= NULL
;
6217 oldsd
->output_queue_tailp
= &oldsd
->output_queue
;
6219 /* Append NAPI poll list from offline CPU. */
6220 if (!list_empty(&oldsd
->poll_list
)) {
6221 list_splice_init(&oldsd
->poll_list
, &sd
->poll_list
);
6222 raise_softirq_irqoff(NET_RX_SOFTIRQ
);
6225 raise_softirq_irqoff(NET_TX_SOFTIRQ
);
6228 /* Process offline CPU's input_pkt_queue */
6229 while ((skb
= __skb_dequeue(&oldsd
->process_queue
))) {
6231 input_queue_head_incr(oldsd
);
6233 while ((skb
= __skb_dequeue(&oldsd
->input_pkt_queue
))) {
6235 input_queue_head_incr(oldsd
);
6243 * netdev_increment_features - increment feature set by one
6244 * @all: current feature set
6245 * @one: new feature set
6246 * @mask: mask feature set
6248 * Computes a new feature set after adding a device with feature set
6249 * @one to the master device with current feature set @all. Will not
6250 * enable anything that is off in @mask. Returns the new feature set.
6252 netdev_features_t
netdev_increment_features(netdev_features_t all
,
6253 netdev_features_t one
, netdev_features_t mask
)
6255 if (mask
& NETIF_F_GEN_CSUM
)
6256 mask
|= NETIF_F_ALL_CSUM
;
6257 mask
|= NETIF_F_VLAN_CHALLENGED
;
6259 all
|= one
& (NETIF_F_ONE_FOR_ALL
|NETIF_F_ALL_CSUM
) & mask
;
6260 all
&= one
| ~NETIF_F_ALL_FOR_ALL
;
6262 /* If one device supports hw checksumming, set for all. */
6263 if (all
& NETIF_F_GEN_CSUM
)
6264 all
&= ~(NETIF_F_ALL_CSUM
& ~NETIF_F_GEN_CSUM
);
6268 EXPORT_SYMBOL(netdev_increment_features
);
6270 static struct hlist_head
*netdev_create_hash(void)
6273 struct hlist_head
*hash
;
6275 hash
= kmalloc(sizeof(*hash
) * NETDEV_HASHENTRIES
, GFP_KERNEL
);
6277 for (i
= 0; i
< NETDEV_HASHENTRIES
; i
++)
6278 INIT_HLIST_HEAD(&hash
[i
]);
6283 /* Initialize per network namespace state */
6284 static int __net_init
netdev_init(struct net
*net
)
6286 if (net
!= &init_net
)
6287 INIT_LIST_HEAD(&net
->dev_base_head
);
6289 net
->dev_name_head
= netdev_create_hash();
6290 if (net
->dev_name_head
== NULL
)
6293 net
->dev_index_head
= netdev_create_hash();
6294 if (net
->dev_index_head
== NULL
)
6300 kfree(net
->dev_name_head
);
6306 * netdev_drivername - network driver for the device
6307 * @dev: network device
6309 * Determine network driver for device.
6311 const char *netdev_drivername(const struct net_device
*dev
)
6313 const struct device_driver
*driver
;
6314 const struct device
*parent
;
6315 const char *empty
= "";
6317 parent
= dev
->dev
.parent
;
6321 driver
= parent
->driver
;
6322 if (driver
&& driver
->name
)
6323 return driver
->name
;
6327 int __netdev_printk(const char *level
, const struct net_device
*dev
,
6328 struct va_format
*vaf
)
6332 if (dev
&& dev
->dev
.parent
)
6333 r
= dev_printk(level
, dev
->dev
.parent
, "%s: %pV",
6334 netdev_name(dev
), vaf
);
6336 r
= printk("%s%s: %pV", level
, netdev_name(dev
), vaf
);
6338 r
= printk("%s(NULL net_device): %pV", level
, vaf
);
6342 EXPORT_SYMBOL(__netdev_printk
);
6344 int netdev_printk(const char *level
, const struct net_device
*dev
,
6345 const char *format
, ...)
6347 struct va_format vaf
;
6351 va_start(args
, format
);
6356 r
= __netdev_printk(level
, dev
, &vaf
);
6361 EXPORT_SYMBOL(netdev_printk
);
6363 #define define_netdev_printk_level(func, level) \
6364 int func(const struct net_device *dev, const char *fmt, ...) \
6367 struct va_format vaf; \
6370 va_start(args, fmt); \
6375 r = __netdev_printk(level, dev, &vaf); \
6380 EXPORT_SYMBOL(func);
6382 define_netdev_printk_level(netdev_emerg
, KERN_EMERG
);
6383 define_netdev_printk_level(netdev_alert
, KERN_ALERT
);
6384 define_netdev_printk_level(netdev_crit
, KERN_CRIT
);
6385 define_netdev_printk_level(netdev_err
, KERN_ERR
);
6386 define_netdev_printk_level(netdev_warn
, KERN_WARNING
);
6387 define_netdev_printk_level(netdev_notice
, KERN_NOTICE
);
6388 define_netdev_printk_level(netdev_info
, KERN_INFO
);
6390 static void __net_exit
netdev_exit(struct net
*net
)
6392 kfree(net
->dev_name_head
);
6393 kfree(net
->dev_index_head
);
6396 static struct pernet_operations __net_initdata netdev_net_ops
= {
6397 .init
= netdev_init
,
6398 .exit
= netdev_exit
,
6401 static void __net_exit
default_device_exit(struct net
*net
)
6403 struct net_device
*dev
, *aux
;
6405 * Push all migratable network devices back to the
6406 * initial network namespace
6409 for_each_netdev_safe(net
, dev
, aux
) {
6411 char fb_name
[IFNAMSIZ
];
6413 /* Ignore unmoveable devices (i.e. loopback) */
6414 if (dev
->features
& NETIF_F_NETNS_LOCAL
)
6417 /* Leave virtual devices for the generic cleanup */
6418 if (dev
->rtnl_link_ops
)
6421 /* Push remaining network devices to init_net */
6422 snprintf(fb_name
, IFNAMSIZ
, "dev%d", dev
->ifindex
);
6423 err
= dev_change_net_namespace(dev
, &init_net
, fb_name
);
6425 pr_emerg("%s: failed to move %s to init_net: %d\n",
6426 __func__
, dev
->name
, err
);
6433 static void __net_exit
default_device_exit_batch(struct list_head
*net_list
)
6435 /* At exit all network devices most be removed from a network
6436 * namespace. Do this in the reverse order of registration.
6437 * Do this across as many network namespaces as possible to
6438 * improve batching efficiency.
6440 struct net_device
*dev
;
6442 LIST_HEAD(dev_kill_list
);
6445 list_for_each_entry(net
, net_list
, exit_list
) {
6446 for_each_netdev_reverse(net
, dev
) {
6447 if (dev
->rtnl_link_ops
)
6448 dev
->rtnl_link_ops
->dellink(dev
, &dev_kill_list
);
6450 unregister_netdevice_queue(dev
, &dev_kill_list
);
6453 unregister_netdevice_many(&dev_kill_list
);
6454 list_del(&dev_kill_list
);
6458 static struct pernet_operations __net_initdata default_device_ops
= {
6459 .exit
= default_device_exit
,
6460 .exit_batch
= default_device_exit_batch
,
6464 * Initialize the DEV module. At boot time this walks the device list and
6465 * unhooks any devices that fail to initialise (normally hardware not
6466 * present) and leaves us with a valid list of present and active devices.
6471 * This is called single threaded during boot, so no need
6472 * to take the rtnl semaphore.
6474 static int __init
net_dev_init(void)
6476 int i
, rc
= -ENOMEM
;
6478 BUG_ON(!dev_boot_phase
);
6480 if (dev_proc_init())
6483 if (netdev_kobject_init())
6486 INIT_LIST_HEAD(&ptype_all
);
6487 for (i
= 0; i
< PTYPE_HASH_SIZE
; i
++)
6488 INIT_LIST_HEAD(&ptype_base
[i
]);
6490 if (register_pernet_subsys(&netdev_net_ops
))
6494 * Initialise the packet receive queues.
6497 for_each_possible_cpu(i
) {
6498 struct softnet_data
*sd
= &per_cpu(softnet_data
, i
);
6500 memset(sd
, 0, sizeof(*sd
));
6501 skb_queue_head_init(&sd
->input_pkt_queue
);
6502 skb_queue_head_init(&sd
->process_queue
);
6503 sd
->completion_queue
= NULL
;
6504 INIT_LIST_HEAD(&sd
->poll_list
);
6505 sd
->output_queue
= NULL
;
6506 sd
->output_queue_tailp
= &sd
->output_queue
;
6508 sd
->csd
.func
= rps_trigger_softirq
;
6514 sd
->backlog
.poll
= process_backlog
;
6515 sd
->backlog
.weight
= weight_p
;
6516 sd
->backlog
.gro_list
= NULL
;
6517 sd
->backlog
.gro_count
= 0;
6522 /* The loopback device is special if any other network devices
6523 * is present in a network namespace the loopback device must
6524 * be present. Since we now dynamically allocate and free the
6525 * loopback device ensure this invariant is maintained by
6526 * keeping the loopback device as the first device on the
6527 * list of network devices. Ensuring the loopback devices
6528 * is the first device that appears and the last network device
6531 if (register_pernet_device(&loopback_net_ops
))
6534 if (register_pernet_device(&default_device_ops
))
6537 open_softirq(NET_TX_SOFTIRQ
, net_tx_action
);
6538 open_softirq(NET_RX_SOFTIRQ
, net_rx_action
);
6540 hotcpu_notifier(dev_cpu_callback
, 0);
6548 subsys_initcall(net_dev_init
);
6550 static int __init
initialize_hashrnd(void)
6552 get_random_bytes(&hashrnd
, sizeof(hashrnd
));
6556 late_initcall_sync(initialize_hashrnd
);